{
"models": [
{
"id": "hyperflow-8",
"name": "HyperFlow",
"steps": 8,
"label": "新 checkpoint",
"rank": 256,
"alpha": 256,
"video_shift": 12,
"audio_shift": 3,
"version": "v1.0",
"sampler": "HyperFlow fixed grid",
"source": "https://drive.google.com/drive/folders/1109tsD-m-y9CzOzp5B9O0GPxkmMDSpbP"
},
{
"id": "lightx2v-8",
"name": "LightX2V Turbo",
"steps": 8,
"label": "8-step 对照",
"rank": 128,
"alpha": 8,
"video_shift": 6,
"audio_shift": 3,
"version": "v1.0 768p",
"sampler": "uniform grid before shift",
"source": "https://huggingface.co/lightx2v/Minimax-h3-Turbo"
},
{
"id": "fasth3-4",
"name": "FastH3",
"steps": 4,
"label": "GitHub 原速度配置",
"rank": 64,
"alpha": 64,
"video_shift": 12,
"audio_shift": 3,
"version": "dense-datafree preview v1",
"sampler": "uniform grid before shift",
"source": "https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-LoRA"
},
{
"id": "lightx2v-4",
"name": "LightX2V Turbo",
"steps": 4,
"label": "4-step 对照",
"rank": 128,
"alpha": 8,
"video_shift": 6,
"audio_shift": 3,
"version": "v1.2 768p",
"sampler": "uniform grid before shift",
"source": "https://huggingface.co/lightx2v/Minimax-h3-Turbo"
}
],
"samples": [
{
"id": 1,
"slug": "via-mlx-serve-by-561487",
"title": {
"en": "via MLX Serve by",
"zh": "小恐龙打喷嚏变出蝴蝶"
},
"category": "animation",
"source": {
"kind": "x",
"name": "@albertgao",
"url": "https://x.com/albertgao/status/2085137397134561487"
},
"source_prompt": "\"A cute little girl walks through a sunny forest when a small, friendly dinosaur steps out, waves, and says hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style.\"\n\n- time: 15min ish\n- 768x768",
"source_duration": "5s",
"prompt": "integrated_multimodal_description: [Shot 1] A cute little girl walks through a sunny forest when a small, friendly dinosaur with a light, cheerful English voice (S1) steps out, waves, and says: [English] hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style. Pace the meeting, sneeze and shared laughter over a continuous 15-second scene, allowing time to see the butterflies settle around both characters.\n\noverall_soundscape: Light forest wind, birds and small footsteps accompany the meeting. A comic sneeze releases a fluttering rush of butterfly wings, followed by the girl and dinosaur laughing together.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "954337465d95f9ab7b1fd512ba3879fa925f8e668f493e314c9d84a91e99fad3",
"duration": 15,
"seed": 42,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 5s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "via MLX Serve by",
"zh": "via MLX Serve by"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 2,
"slug": "the-last-thing-you-see-in-your-first-and-last-290296",
"title": {
"en": "The last thing you see in your first and last",
"zh": "太空旅客窗外的黑洞"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@ivanfioravanti",
"url": "https://x.com/ivanfioravanti/status/2086553101029290296"
},
"source_prompt": "amateur handheld pov footage of a tourist in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: \"Wow look at that!\", then they show back the room",
"source_duration": "10s",
"prompt": "integrated_multimodal_description: [Shot 1] amateur handheld pov footage of a tourist with an excited, conversational English voice (S1) in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: [English] Wow look at that!, then they show back the room Use one uninterrupted 15-second handheld take. Switch off the room light around halfway through, hold the window view for the spoken reaction, then turn back to show the room.\n\noverall_soundscape: A steady enclosed-room ventilation hum and soft carpeted footsteps continue under slight handheld handling noise. A light-switch click marks the change to the darkened room.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "712f02d91f39b4d5a029cf4c3396609ac32d151511bf04949501f697139405e5",
"duration": 15,
"seed": 43,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 10s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "The last thing you see in your first and last",
"zh": "The last thing you see in your first and last"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 3,
"slug": "ultra-cinematic-macro-shot-a-calm-mountain-lake-039920",
"title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "水滴汇成 MORNING"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2085622021812039920"
},
"source_prompt": "Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality. In one continuous 15-second camera move, establish the dawn lake first, raise and spiral the droplets in the middle, hold the completed readable word before its final liquid collapse.\n\noverall_soundscape: Quiet mountain wind, small lake ripples and faint distant dawn birds establish the lakeside. Delicate water movement grows into the rushing liquid-letter collapse and a crisp shower of droplets.\n\nnon_diegetic_music: Slow, sustained string notes and sparse high piano tones rise gently with the water lettering and resolve as the droplets fall.",
"prompt_sha256": "7abb51ba6a7327cb9266b13da9a8fa218928aeb0d97f444c82dc95ea21c245c2",
"duration": 15,
"seed": 44,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "Ultra cinematic macro shot: A calm mountain lake at dawn"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 4,
"slug": "strawberry-seasonal-match-cut-food-commercial-117943",
"title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓穿过四季落入奶油"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@HBCoop_",
"url": "https://x.com/HBCoop_/status/2082827829172117943"
},
"source_prompt": "Premium food commercial. Extreme macro shot of a single ripe strawberry falling toward a white ceramic bowl. Before it lands, the environment changes through four seamless seasonal match cuts: a sunlit spring greenhouse, a hot summer picnic, an autumn kitchen and a candlelit winter dining room. The strawberry stays in exactly the same screen position, size, orientation and downward motion through every transition. It lands in fresh cream during the winter scene, sending one graceful crown-shaped splash upward. Photorealistic food texture, high-speed product cinematography, clean commercial lighting. Each season has distinct environmental sound, but one continuous musical phrase connects the entire sequence. No hands, labels or additional fruit.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Photorealistic premium food commercial, extreme macro high-speed product cinematography. A single ripe strawberry falls toward a white ceramic bowl in a sunlit spring greenhouse. Its red seeds and moist surface are sharply resolved against soft green glasshouse bokeh. Keep the strawberry's screen position, size, orientation and downward motion consistent through every seasonal match cut. No hands, labels or additional fruit.\n[Shot 2] At 00:03.000, the shot transitions to a hot summer picnic in a seamless seasonal match cut around the same falling strawberry and bowl. Warm sunlight replaces the greenhouse light; distant insects accompany its uninterrupted descent.\n[Shot 3] At 00:06.000, the shot transitions to an autumn kitchen with amber window light, preserving the strawberry and bowl alignment and the same downward motion. Soft indoor room tone replaces the outdoor insects.\n[Shot 4] At 00:09.000, the shot transitions to a candlelit winter dining room. The strawberry completes its fall into fresh cream in the white bowl, sending one graceful crown-shaped splash upward. Hold the detailed splash and its settling ripples through the final seconds. One continuous musical phrase connects all four seasons.\n\noverall_soundscape: Spring birds and a gentle greenhouse breeze change to summer insects, quiet autumn kitchen ambience, then the still winter dining room. The final cream splash and falling droplets remain clearly audible.\n\nnon_diegetic_music: One uninterrupted light piano-and-plucked-string phrase runs at a moderate tempo across all four seasons, resolving with the strawberry splash.",
"prompt_sha256": "0c4614d6b1c88a86974b1393eab7a574817abc11c37bf72e3cb53e2d9bbf718f",
"duration": 15,
"seed": 45,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓四季匹配剪辑食品广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 5,
"slug": "stunning-action-packed-scene-following-a-little-robot-r-288619",
"title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "赛博城市里的机器人追逐"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@AIandDesign",
"url": "https://x.com/AIandDesign/status/2082522979339288619"
},
"source_prompt": "A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown. Maintain one continuous 15-second rear chase shot with brief arcs that reveal the pursuers when the robot looks back; preserve the robot’s identity and forward motion.\n\noverall_soundscape: Rapid metallic footfalls and small robot servo movements run through the city ambience. Heavy pursuing footsteps, drone rotors and passing machinery track the chase without intelligible dialogue.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be75f61112b0d84692fb5ce91ced32ae50c3f1a3a1ad6b115fba2089944d4073",
"duration": 15,
"seed": 46,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "小机器人赛博朋克逃亡"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 6,
"slug": "created-with-minimax-h3-max-120666",
"title": {
"en": "Created with MiniMax H3 Max",
"zh": "湖畔花束与风"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@TaliaAariz",
"url": "https://x.com/TaliaAariz/status/2095857062987120666"
},
"source_prompt": "A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field. Keep one continuous 15-second medium shot, with a very slow small push-in as she touches the flowers, smiles and finally closes her eyes to feel the breeze.\n\noverall_soundscape: Gentle shore wind moves the ribbons and reeds while small waves wash over the pebbles. Quiet bouquet and fabric rustles accompany her hand movements.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7e05bce3aa444d99e753af9bcc4865bd2934e2c9102f5a1ce63f952302516d15",
"duration": 15,
"seed": 47,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3 Max",
"zh": "Created with MiniMax H3 Max"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 7,
"slug": "created-with-minimax-h3-on-989121",
"title": {
"en": "Created with MiniMax H3 on .",
"zh": "画家与湖畔城市天际线"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087408671101989121"
},
"source_prompt": "A young Western artist sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western artist with a relaxed conversational English voice (S1) sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience. Her brief natural English remarks follow her painting actions. Keep the artist, easel and skyline consistent over the 15-second take. Begin with her positioning the easel, follow several visible finishing brush strokes, then give the final canvas-to-skyline comparison time to read.\n\noverall_soundscape: Soft city-park wind, distant traffic and lakeside birds surround the easel. Light wood movements, brush-on-canvas friction and cloth handling follow the artist’s actions.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "f39be188be70b922d6337aa1436e276c98592d5adf01475408b43bbd227e6cbc",
"duration": 15,
"seed": 48,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Created with MiniMax H3 on .",
"zh": "Created with MiniMax H3 on ."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 8,
"slug": "created-with-minimax-h3-522089",
"title": {
"en": "Created with MiniMax H3.",
"zh": "街头摄影师、老人和小狗"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087102541146522089"
},
"source_prompt": "A young Western female street photographer walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says “Look at that,” then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western female street photographer with a warm conversational English voice (S1) walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says: [English] Look at that, then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism. Over one continuous 15-second handheld shot, follow her noticing and photographing the café scene, then turn with her as she shows the camera display, speaks and continues walking.\n\noverall_soundscape: Layered downtown footsteps, distant traffic and soft café room spill continue beneath a camera shutter click. The dog shifts and lightly pants beside the seated man.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be91933cc24647e7105ff883517384c8c83c2ccef42801f72555f509c467e319",
"duration": 15,
"seed": 49,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3.",
"zh": "Created with MiniMax H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 9,
"slug": "low-angle-fashion-tracking-film-062019",
"title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "小巷里的时装与歌声"
},
"category": "fashion",
"source": {
"kind": "x",
"name": "@Kiber_Alla",
"url": "https://x.com/Kiber_Alla/status/2083583963512062019"
},
"source_prompt": "Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. She sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. The woman, with a light melodic singing voice (S1), sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look. She performs a wordless vocal melody. Keep the low-angle backward tracking shot continuous for the full 15 seconds. Leave enough space for a full turn and return to eye contact without cutting.\n\noverall_soundscape: Light footsteps on stone, wind through the alley and natural fabric movement accompany the performance. Her soft laughter and breaths are synchronized to the visible movement.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7c2058cc5dc84a848b29169d61ecc607158e3ce4e68dd07d43622c10bfd34d91",
"duration": 15,
"seed": 50,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词或歌词；未编造逐字台词，歌唱使用无词旋律。"
],
"source_title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "低机位时尚跟拍短片"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 10,
"slug": "generated-a-french-bulldog-with-minimax-h3-182507",
"title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "法国斗牛犬闯障碍"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085448660612182507"
},
"source_prompt": "A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose. Follow the complete obstacle sequence in one 15-second handheld take, with the bell contact visible in the final seconds.\n\noverall_soundscape: Backyard birds and distant neighborhood ambience surround quick paw taps and excited dog panting. Plastic poles rattle, the fabric tunnel scrapes, and a small bell rings once at the finish.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "b441f7d70cc0c7a54efccb096492e34839955225bc31781700cf560af438bc8d",
"duration": 15,
"seed": 51,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "Generated a french bulldog with MiniMax H3"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 11,
"slug": "created-with-hailuo-h3-182619",
"title": {
"en": "Created with Hailuo H3.",
"zh": "冰川崩塌组成 EXTINCTION"
},
"category": "title-sequence",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2090740724803182619"
},
"source_prompt": "Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX. Over 15 seconds, establish the glacier, follow the expanding fractures and airborne ice, hold the completed readable ice title, then finish on its violent shatter.\n\noverall_soundscape: Deep frozen-ocean wind surrounds spreading ice creaks and resonant fractures. Heavy ice collapse becomes a crystalline debris storm, followed by the sharp final title shatter.\n\nnon_diegetic_music: Sustained low strings swell with the ice fracture, joined by deep percussion at the shelf collapse and final title shatter.",
"prompt_sha256": "4450f406537c05f4eaf4263bdbb5e1e4d60f9b8e253dbe76aac8d0be5e2b24bd",
"duration": 15,
"seed": 52,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with Hailuo H3.",
"zh": "Created with Hailuo H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 12,
"slug": "shatter-695817",
"title": {
"en": "SHATTER.",
"zh": "玻璃碎片组成 SHATTER"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2096554023809695817"
},
"source_prompt": "Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX. Keep the camera move continuous over 15 seconds: establish the floating monolith, enter the spreading fracture, orbit the suspended fragments, hold the completed title briefly, then finish on fragments expanding toward the lens.\n\noverall_soundscape: A projectile impact triggers fine traveling glass cracks and a swelling structural groan. Dense crystalline shattering and passing shards follow the orbit, with a second pressure-wave crack and outward burst at the end.\n\nnon_diegetic_music: A sparse low electronic pulse builds beneath bowed-metal tones, with two deep percussion hits at the main and final glass explosions.",
"prompt_sha256": "b198e1ddb0d1d5c79cab984ce06e7944ecd4efbca41f105960c7312c540dd572",
"duration": 15,
"seed": 53,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "SHATTER.",
"zh": "SHATTER."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 13,
"slug": "shot-opens-tight-on-a-beautiful-woman-face-wind-452380",
"title": {
"en": "Cliffside Descent to the Ocean",
"zh": "从悬崖人像到俯拍海浪"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@umesh_ai",
"url": "https://x.com/umesh_ai/status/2082700637444452380"
},
"source_prompt": "The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below. Perform the described camera path as one continuous 15-second move, ending on the high overhead view of the lone figure and exploding surf.\n\noverall_soundscape: Strong coastal wind drives the coat fabric. Deep ocean roar becomes clearer as the camera descends, with violent waves striking rock beneath the final overhead view.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "8a9fb882646cdb521b55ea777eea72c255e9b3863f257602d34d5bda4da3b879",
"duration": 15,
"seed": 54,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Cliffside Descent to the Ocean",
"zh": "悬崖下行至海边"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 14,
"slug": "ugc-style-skincare-video-featuring-a-realistic-young-wo-242986",
"title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "镜前护肤产品口播"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@ZaraIrahh",
"url": "https://x.com/ZaraIrahh/status/2083011066800242986"
},
"source_prompt": "UGC-style skincare video featuring a realistic young woman speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, vertical 9:16, social media ad quality, highly realistic natural skin texture, not overly retouched.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] UGC-style skincare video featuring a realistic young woman with a clear conversational English voice (S1) speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, landscape 16:9, social media ad quality, highly realistic natural skin texture, not overly retouched. Her short natural English presentation stays about her skincare routine. Use a continuous 15-second landscape smartphone shot framed to keep both her face and the bottle clearly visible. She presents the serum, applies it to her face, and returns her attention to the camera.\n\noverall_soundscape: Quiet bathroom room tone, small glass-bottle handling sounds and soft hand-on-skin movements accompany the presentation. Keep breaths and clothing movement natural and close.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "61f30fc7218593117f92f6c7e46d55e03de10b31642a4122a69a9eba43d40b4c",
"duration": 15,
"seed": 55,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "浴室镜前护肤 UGC 广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 15,
"slug": "found-footage-film-with-335284",
"title": {
"en": "Horror Film Study 335284",
"zh": "渔夫接到诡异来电"
},
"category": "horror",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2084842607411335284"
},
"source_prompt": "\"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A distorted voice says, \"Finally. I’ve been trying to switch places with you.\" The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A low distorted English voice from the phone speaker (S1) says: [English] Finally. I’ve been trying to switch places with you. The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\" Keep the cheap boat-mounted camera viewpoint continuous for the 15-second scene, with only the specified white-flash glitch interrupting the image. Establish the fishing action, allow the phone line to finish, show the trapped image and then the wet double before the abrupt final black frame.\n\noverall_soundscape: Low lake water lapping, boat creaks and fishing-reel strain accompany the fisherman’s heavy breathing. Wet weeds scrape the hull, the phone glitches with clipped electronic noise, then water splashes as the soaked double enters the boat.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "778e3b3e9a6ca16d59312279338db624dd6281b72e5ade48402f63ed4b25f617",
"duration": 15,
"seed": 56,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Horror Film Study 335284",
"zh": "恐怖短片案例 335284"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 16,
"slug": "cinematic-sci-fi-suspense-video-a-realistic-spaceship-crew-in-024495",
"title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "太空船门后的小学教室"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085498501384024495"
},
"source_prompt": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic uniforms follows a mysterious distress signal and boards a dark, abandoned spacecraft drifting in space. Flashlights cut through darkness as they move cautiously through narrow metallic corridors. Warning lights flicker, metal creaks, and their boots echo on the floor. One crew member says, \"Signal's coming from deeper inside.\" They enter a large chamber filled with hundreds of empty cryogenic pods, all open and abandoned. Another crew member says nervously, \"Where did everyone go?\" They approach a final closed door at the far end. The captain slowly opens it, and instead of another ship compartment, it reveals a completely normal present-day elementary school classroom in bright daylight. Children sit at desks, backpacks on the floor, colorful posters on the walls, and a teacher at the front turns to the crew and scolds them: \"You're late again. Take your seats.\" The astronauts stand frozen in confusion at the doorway. Realistic tone, smooth camera movement, strong suspense buildup, with the final reveal clear, sudden, and absurd. Include realistic audio: radio static, low ship hum, footsteps, pod machinery ambience, door hiss, then ordinary classroom sounds and the teacher's voice.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic science-fiction suspense, realistic spaceship crew in full futuristic uniforms boarding a dark abandoned spacecraft in response to a distress signal. A smooth tracking shot follows them through a narrow metallic corridor; flashlight beams cross flickering warning lights. One crew member with a restrained mid-pitched English voice (S1) says: [English] Signal's coming from deeper inside. Radio static, a low ship hum, creaking metal and hard boot echoes accompany their advance.\n[Shot 2] At 00:04.500, the camera cuts to a wide view of a large chamber with hundreds of open, empty cryogenic pods. A second crew member with an uneasy, slightly higher English voice (S2) looks from the abandoned pods toward a final closed door and says nervously: [English] Where did everyone go? The captain approaches that door as the camera pushes forward with the crew. Pod machinery resonates beneath their footsteps.\n[Shot 3] At 00:09.000, the camera cuts behind the captain as the door hisses open. Bright daylight reveals an ordinary present-day elementary school classroom, children at desks, backpacks on the floor and colorful posters on the walls. A teacher at the front with a clear, matter-of-fact English voice (S3) turns toward the astronauts and scolds them: [English] You're late again. Take your seats. The astronauts remain frozen in confusion at the doorway. Hold the absurd reveal until the end while ordinary classroom room tone replaces the ship ambience.\n\noverall_soundscape: Radio static, a low ship hum, creaking metal, boot echoes and empty pod machinery fill the spacecraft. A door hiss gives way abruptly to ordinary classroom chair movement and quiet paper rustling.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "515b6f9c02a9aff5ebdd48c593434b27848f0a398afc1c0651103dfdd475c0da",
"duration": 15,
"seed": 57,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 17,
"slug": "skyship-flight-across-a-floating-kingdom-single-continu-344891",
"title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "穿越浮岛王国的飞艇"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@Strength04_X",
"url": "https://x.com/Strength04_X/status/2082712692159344891"
},
"source_prompt": "Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds. Complete the described path in one uninterrupted 15-second flight, establishing the skyship first and reserving the final seconds for the sunlight reveal beyond the storm.\n\noverall_soundscape: High-altitude wind and engine thrust deepen as the skyship enters thunder, lashing rain and cracking rock. Waterfalls rush past the hull before the storm sound falls behind the ship in open sunlight.\n\nnon_diegetic_music: Broad sustained strings and steady low orchestral percussion build through the storm, then open into a long brass-and-string chord in sunlight.",
"prompt_sha256": "15ed03a4ff9ad4d170616c8220d15f8c7ddadb16a059a522c567a1cede541838",
"duration": 15,
"seed": 58,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "飞空艇穿越浮空王国"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 18,
"slug": "a-hyper-realistic-handheld-phone-video-of-a-quiet-suburban-495564",
"title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "后院上空突然坠落的天空"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2086879654116495564"
},
"source_prompt": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\" Use a single 15-second home-video take. Preserve the first several seconds of ordinary watering, begin the overhead crack after that peaceful lead-in, and end with the sky descending and camera falling.\n\noverall_soundscape: Ordinary backyard birds, distant traffic and a watering stream continue for several seconds. A sudden deep overhead crack becomes a heavy descending rumble, followed by startled breathing and the phone hitting the ground.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "42f8695e30223aeca06566bf0476f7882751c30d3db90ad0136585d517c6ba26",
"duration": 15,
"seed": 59,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 19,
"slug": "also-has-an-impressive-understanding-of-differen-911311",
"title": {
"en": "also has an impressive understanding of different cultures",
"zh": "穿行爪哇市场的肉丸摊贩"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@junwatu",
"url": "https://x.com/junwatu/status/2084840152715911311"
},
"source_prompt": "Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism. Nearby women bargaining in Javanese (S1) and vendors calling prices in Javanese (S2) remain overlapping background voices. Follow the vendor in one continuous 15-second handheld take, keeping the steaming bowl and his careful steps legible; finish with serving the bowl and the brief power flicker.\n\noverall_soundscape: Dense covered-market bustle includes footsteps, bowl clatter, steaming food and sacks brushing clothing. Slow motorcycle engines pass among the stalls, followed by a brief fluorescent power buzz and the uninterrupted market activity.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "644ee708e360b5247f1f86b9510bc54be46b1ede81be1dfd8e283227dfab466f",
"duration": 15,
"seed": 60,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "also has an impressive understanding of different cultures",
"zh": "also has an impressive understanding of different cultures"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
},
{
"id": 20,
"slug": "16-9-cinematic-wuxia-mystery-set-in-a-bamboo-571041",
"title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "夜雪竹林里的武侠悬疑"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@AIwithAliya",
"url": "https://x.com/AIwithAliya/status/2083132770650571041"
},
"source_prompt": "A 16:9 cinematic wuxia mystery set in a bamboo forest at night. Use a low-saturation palette of cold blue, ink green, charcoal, and gray. Thin mist fills the forest and fine snow drifts through the air. The mood is austere, lethal, and controlled, with the tension of a martial-arts sect investigating a case and exchanging secret intelligence.\n\nDeep in the forest, dense vertical bamboo fills the background while cold white mist-light glows in the distance. Soft, out-of-focus leaves partially obscure the foreground, creating the sense of watching from within the grove. A cool, soft front-side key lights the actors’ faces; backlight keeps the foreground dark and the distance luminous. Use shallow depth of field so leaves, snow, and bamboo dissolve into soft bokeh.\nPrioritize facial close-ups and measured shot/reverse-shot coverage. Keep the rhythm restrained but tense. Photoreal period-drama production value, cinematic lighting, and no modern elements. No subtitles, on-screen text, watermarks, modern clothing or architecture, animation styling, over-smoothed skin, bright daylight, or comic performance.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic wuxia mystery in a bamboo forest at night, in a restrained cold-blue, ink-green, charcoal and gray palette. Thin mist and fine drifting snow surround two period-clothed martial-arts investigators. Dense vertical bamboo fills the background; soft foreground leaves partially obscure a tight facial close-up of the first investigator. Cool, soft front-side light reveals natural skin texture while cold white mist-light glows in the distance. The investigator looks toward the other person with measured, wary attention. No modern objects, text, subtitles, watermarks or comic performance.\n[Shot 2] At 00:05.000, the camera cuts to the other investigator in matching close-up for a restrained reverse angle. Keep the same cold lighting, shallow depth of field and softly blurred falling snow. A small shift of the eyes toward the dark grove conveys the exchange of secret intelligence without adding invented dialogue. Clothing rustles quietly under the bamboo wind.\n[Shot 3] At 00:10.000, the camera cuts back to the first investigator seen through out-of-focus leaves. A slow, small push-in follows a controlled glance into the luminous mist. Hold the austere, tense expression until the end, with the distant forest remaining still except for snow and moving leaves. Photoreal period-drama lighting, no animation styling, bright daylight or over-smoothed skin.\n\noverall_soundscape: Thin wind passes through bamboo leaves with occasional dry stalk creaks. Soft period-clothing movement and faint footsteps remain close while falling snow and distant forest space stay quiet.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "11628027721bd267f57680bb10ff37aff3623b6ea6bbfb064178144173e3a828",
"duration": 15,
"seed": 61,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "竹林武侠谜案"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt"
}
],
"outputs": [
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 1,
"slug": "via-mlx-serve-by-561487",
"title": {
"en": "via MLX Serve by",
"zh": "小恐龙打喷嚏变出蝴蝶"
},
"category": "animation",
"source": {
"kind": "x",
"name": "@albertgao",
"url": "https://x.com/albertgao/status/2085137397134561487"
},
"source_prompt": "\"A cute little girl walks through a sunny forest when a small, friendly dinosaur steps out, waves, and says hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style.\"\n\n- time: 15min ish\n- 768x768",
"source_duration": "5s",
"prompt": "integrated_multimodal_description: [Shot 1] A cute little girl walks through a sunny forest when a small, friendly dinosaur with a light, cheerful English voice (S1) steps out, waves, and says: [English] hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style. Pace the meeting, sneeze and shared laughter over a continuous 15-second scene, allowing time to see the butterflies settle around both characters.\n\noverall_soundscape: Light forest wind, birds and small footsteps accompany the meeting. A comic sneeze releases a fluttering rush of butterfly wings, followed by the girl and dinosaur laughing together.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "954337465d95f9ab7b1fd512ba3879fa925f8e668f493e314c9d84a91e99fad3",
"duration": 15,
"seed": 42,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 5s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "via MLX Serve by",
"zh": "via MLX Serve by"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.019709377083927,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/01.mp4",
"poster": "assets/v2-15s/fasth3-4/01.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6482287,
"sha256": "6541ab0ed98b8e1c74c21ca8be0e11d000c9577f519d91484def8aa4a66edeb2",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/01.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/01.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 2,
"slug": "the-last-thing-you-see-in-your-first-and-last-290296",
"title": {
"en": "The last thing you see in your first and last",
"zh": "太空旅客窗外的黑洞"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@ivanfioravanti",
"url": "https://x.com/ivanfioravanti/status/2086553101029290296"
},
"source_prompt": "amateur handheld pov footage of a tourist in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: \"Wow look at that!\", then they show back the room",
"source_duration": "10s",
"prompt": "integrated_multimodal_description: [Shot 1] amateur handheld pov footage of a tourist with an excited, conversational English voice (S1) in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: [English] Wow look at that!, then they show back the room Use one uninterrupted 15-second handheld take. Switch off the room light around halfway through, hold the window view for the spoken reaction, then turn back to show the room.\n\noverall_soundscape: A steady enclosed-room ventilation hum and soft carpeted footsteps continue under slight handheld handling noise. A light-switch click marks the change to the darkened room.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "712f02d91f39b4d5a029cf4c3396609ac32d151511bf04949501f697139405e5",
"duration": 15,
"seed": 43,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 10s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "The last thing you see in your first and last",
"zh": "The last thing you see in your first and last"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.38888839399442,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/02.mp4",
"poster": "assets/v2-15s/fasth3-4/02.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6844578,
"sha256": "1206f7816cd4c0b865d28738fead5c660828eb90406bdea3f1832c4ab82558a4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/02.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/02.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 3,
"slug": "ultra-cinematic-macro-shot-a-calm-mountain-lake-039920",
"title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "水滴汇成 MORNING"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2085622021812039920"
},
"source_prompt": "Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality. In one continuous 15-second camera move, establish the dawn lake first, raise and spiral the droplets in the middle, hold the completed readable word before its final liquid collapse.\n\noverall_soundscape: Quiet mountain wind, small lake ripples and faint distant dawn birds establish the lakeside. Delicate water movement grows into the rushing liquid-letter collapse and a crisp shower of droplets.\n\nnon_diegetic_music: Slow, sustained string notes and sparse high piano tones rise gently with the water lettering and resolve as the droplets fall.",
"prompt_sha256": "7abb51ba6a7327cb9266b13da9a8fa218928aeb0d97f444c82dc95ea21c245c2",
"duration": 15,
"seed": 44,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "Ultra cinematic macro shot: A calm mountain lake at dawn"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 17.910059693735093,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/03.mp4",
"poster": "assets/v2-15s/fasth3-4/03.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 10485091,
"sha256": "2035c6a3eb0046e08b41001ff250eac42d80bd117e9e8c7fc240cd25ce08a133",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/03.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/03.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 4,
"slug": "strawberry-seasonal-match-cut-food-commercial-117943",
"title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓穿过四季落入奶油"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@HBCoop_",
"url": "https://x.com/HBCoop_/status/2082827829172117943"
},
"source_prompt": "Premium food commercial. Extreme macro shot of a single ripe strawberry falling toward a white ceramic bowl. Before it lands, the environment changes through four seamless seasonal match cuts: a sunlit spring greenhouse, a hot summer picnic, an autumn kitchen and a candlelit winter dining room. The strawberry stays in exactly the same screen position, size, orientation and downward motion through every transition. It lands in fresh cream during the winter scene, sending one graceful crown-shaped splash upward. Photorealistic food texture, high-speed product cinematography, clean commercial lighting. Each season has distinct environmental sound, but one continuous musical phrase connects the entire sequence. No hands, labels or additional fruit.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Photorealistic premium food commercial, extreme macro high-speed product cinematography. A single ripe strawberry falls toward a white ceramic bowl in a sunlit spring greenhouse. Its red seeds and moist surface are sharply resolved against soft green glasshouse bokeh. Keep the strawberry's screen position, size, orientation and downward motion consistent through every seasonal match cut. No hands, labels or additional fruit.\n[Shot 2] At 00:03.000, the shot transitions to a hot summer picnic in a seamless seasonal match cut around the same falling strawberry and bowl. Warm sunlight replaces the greenhouse light; distant insects accompany its uninterrupted descent.\n[Shot 3] At 00:06.000, the shot transitions to an autumn kitchen with amber window light, preserving the strawberry and bowl alignment and the same downward motion. Soft indoor room tone replaces the outdoor insects.\n[Shot 4] At 00:09.000, the shot transitions to a candlelit winter dining room. The strawberry completes its fall into fresh cream in the white bowl, sending one graceful crown-shaped splash upward. Hold the detailed splash and its settling ripples through the final seconds. One continuous musical phrase connects all four seasons.\n\noverall_soundscape: Spring birds and a gentle greenhouse breeze change to summer insects, quiet autumn kitchen ambience, then the still winter dining room. The final cream splash and falling droplets remain clearly audible.\n\nnon_diegetic_music: One uninterrupted light piano-and-plucked-string phrase runs at a moderate tempo across all four seasons, resolving with the strawberry splash.",
"prompt_sha256": "0c4614d6b1c88a86974b1393eab7a574817abc11c37bf72e3cb53e2d9bbf718f",
"duration": 15,
"seed": 45,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓四季匹配剪辑食品广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 16.10258880071342,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/04.mp4",
"poster": "assets/v2-15s/fasth3-4/04.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4354758,
"sha256": "e46ad094d1dc471b260948562878526824f1475bb34ac6d4476b141a217f7d9a",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/04.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/04.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 5,
"slug": "stunning-action-packed-scene-following-a-little-robot-r-288619",
"title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "赛博城市里的机器人追逐"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@AIandDesign",
"url": "https://x.com/AIandDesign/status/2082522979339288619"
},
"source_prompt": "A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown. Maintain one continuous 15-second rear chase shot with brief arcs that reveal the pursuers when the robot looks back; preserve the robot’s identity and forward motion.\n\noverall_soundscape: Rapid metallic footfalls and small robot servo movements run through the city ambience. Heavy pursuing footsteps, drone rotors and passing machinery track the chase without intelligible dialogue.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be75f61112b0d84692fb5ce91ced32ae50c3f1a3a1ad6b115fba2089944d4073",
"duration": 15,
"seed": 46,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "小机器人赛博朋克逃亡"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.840128422249109,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/05.mp4",
"poster": "assets/v2-15s/fasth3-4/05.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 11493828,
"sha256": "53331ff4e26ca80e2abc8d3a063d5be56160a38a1866906f0385c27bb3a3df95",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/05.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/05.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 6,
"slug": "created-with-minimax-h3-max-120666",
"title": {
"en": "Created with MiniMax H3 Max",
"zh": "湖畔花束与风"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@TaliaAariz",
"url": "https://x.com/TaliaAariz/status/2095857062987120666"
},
"source_prompt": "A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field. Keep one continuous 15-second medium shot, with a very slow small push-in as she touches the flowers, smiles and finally closes her eyes to feel the breeze.\n\noverall_soundscape: Gentle shore wind moves the ribbons and reeds while small waves wash over the pebbles. Quiet bouquet and fabric rustles accompany her hand movements.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7e05bce3aa444d99e753af9bcc4865bd2934e2c9102f5a1ce63f952302516d15",
"duration": 15,
"seed": 47,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3 Max",
"zh": "Created with MiniMax H3 Max"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 20.086923273745924,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/06.mp4",
"poster": "assets/v2-15s/fasth3-4/06.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6770280,
"sha256": "da8a01f0702c2765cb7a26a2f08c10e188495570c29a6bcbdecfc9b241322b79",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/06.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/06.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 7,
"slug": "created-with-minimax-h3-on-989121",
"title": {
"en": "Created with MiniMax H3 on .",
"zh": "画家与湖畔城市天际线"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087408671101989121"
},
"source_prompt": "A young Western artist sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western artist with a relaxed conversational English voice (S1) sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience. Her brief natural English remarks follow her painting actions. Keep the artist, easel and skyline consistent over the 15-second take. Begin with her positioning the easel, follow several visible finishing brush strokes, then give the final canvas-to-skyline comparison time to read.\n\noverall_soundscape: Soft city-park wind, distant traffic and lakeside birds surround the easel. Light wood movements, brush-on-canvas friction and cloth handling follow the artist’s actions.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "f39be188be70b922d6337aa1436e276c98592d5adf01475408b43bbd227e6cbc",
"duration": 15,
"seed": 48,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Created with MiniMax H3 on .",
"zh": "Created with MiniMax H3 on ."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 16.880621903110296,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/07.mp4",
"poster": "assets/v2-15s/fasth3-4/07.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6960548,
"sha256": "cbe28a1d7749470547f2feef0e2f475d1c0b7c77f49c50ea00c4d906593b7a0f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/07.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/07.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 8,
"slug": "created-with-minimax-h3-522089",
"title": {
"en": "Created with MiniMax H3.",
"zh": "街头摄影师、老人和小狗"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087102541146522089"
},
"source_prompt": "A young Western female street photographer walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says “Look at that,” then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western female street photographer with a warm conversational English voice (S1) walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says: [English] Look at that, then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism. Over one continuous 15-second handheld shot, follow her noticing and photographing the café scene, then turn with her as she shows the camera display, speaks and continues walking.\n\noverall_soundscape: Layered downtown footsteps, distant traffic and soft café room spill continue beneath a camera shutter click. The dog shifts and lightly pants beside the seated man.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be91933cc24647e7105ff883517384c8c83c2ccef42801f72555f509c467e319",
"duration": 15,
"seed": 49,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3.",
"zh": "Created with MiniMax H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.759661377873272,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/08.mp4",
"poster": "assets/v2-15s/fasth3-4/08.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 14637787,
"sha256": "ecdf783c52abc0773192c9be51a0ae25e5579c7dd4de1db17ca0f33724785f9f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/08.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/08.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 9,
"slug": "low-angle-fashion-tracking-film-062019",
"title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "小巷里的时装与歌声"
},
"category": "fashion",
"source": {
"kind": "x",
"name": "@Kiber_Alla",
"url": "https://x.com/Kiber_Alla/status/2083583963512062019"
},
"source_prompt": "Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. She sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. The woman, with a light melodic singing voice (S1), sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look. She performs a wordless vocal melody. Keep the low-angle backward tracking shot continuous for the full 15 seconds. Leave enough space for a full turn and return to eye contact without cutting.\n\noverall_soundscape: Light footsteps on stone, wind through the alley and natural fabric movement accompany the performance. Her soft laughter and breaths are synchronized to the visible movement.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7c2058cc5dc84a848b29169d61ecc607158e3ce4e68dd07d43622c10bfd34d91",
"duration": 15,
"seed": 50,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词或歌词；未编造逐字台词，歌唱使用无词旋律。"
],
"source_title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "低机位时尚跟拍短片"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.1059017656371,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/09.mp4",
"poster": "assets/v2-15s/fasth3-4/09.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 11326646,
"sha256": "8de629a836f83a636b3d5f822fa8e83260ecfd1204d6b274870e6ced79dc0282",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/09.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/09.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 10,
"slug": "generated-a-french-bulldog-with-minimax-h3-182507",
"title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "法国斗牛犬闯障碍"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085448660612182507"
},
"source_prompt": "A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose. Follow the complete obstacle sequence in one 15-second handheld take, with the bell contact visible in the final seconds.\n\noverall_soundscape: Backyard birds and distant neighborhood ambience surround quick paw taps and excited dog panting. Plastic poles rattle, the fabric tunnel scrapes, and a small bell rings once at the finish.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "b441f7d70cc0c7a54efccb096492e34839955225bc31781700cf560af438bc8d",
"duration": 15,
"seed": 51,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "Generated a french bulldog with MiniMax H3"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.046641102060676,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/10.mp4",
"poster": "assets/v2-15s/fasth3-4/10.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 13964920,
"sha256": "ff61266928916f5dd70406cefdb1a54d8a2ab6bc1fd6a6d8f7bc700443a71f5c",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/10.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/10.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 11,
"slug": "created-with-hailuo-h3-182619",
"title": {
"en": "Created with Hailuo H3.",
"zh": "冰川崩塌组成 EXTINCTION"
},
"category": "title-sequence",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2090740724803182619"
},
"source_prompt": "Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX. Over 15 seconds, establish the glacier, follow the expanding fractures and airborne ice, hold the completed readable ice title, then finish on its violent shatter.\n\noverall_soundscape: Deep frozen-ocean wind surrounds spreading ice creaks and resonant fractures. Heavy ice collapse becomes a crystalline debris storm, followed by the sharp final title shatter.\n\nnon_diegetic_music: Sustained low strings swell with the ice fracture, joined by deep percussion at the shelf collapse and final title shatter.",
"prompt_sha256": "4450f406537c05f4eaf4263bdbb5e1e4d60f9b8e253dbe76aac8d0be5e2b24bd",
"duration": 15,
"seed": 52,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with Hailuo H3.",
"zh": "Created with Hailuo H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.46429255977273,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/11.mp4",
"poster": "assets/v2-15s/fasth3-4/11.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8081341,
"sha256": "924c1eb52a9acb27a09a2e9615d51fa06e3a609ae0f18650e8742f151878e3ec",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/11.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/11.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 12,
"slug": "shatter-695817",
"title": {
"en": "SHATTER.",
"zh": "玻璃碎片组成 SHATTER"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2096554023809695817"
},
"source_prompt": "Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX. Keep the camera move continuous over 15 seconds: establish the floating monolith, enter the spreading fracture, orbit the suspended fragments, hold the completed title briefly, then finish on fragments expanding toward the lens.\n\noverall_soundscape: A projectile impact triggers fine traveling glass cracks and a swelling structural groan. Dense crystalline shattering and passing shards follow the orbit, with a second pressure-wave crack and outward burst at the end.\n\nnon_diegetic_music: A sparse low electronic pulse builds beneath bowed-metal tones, with two deep percussion hits at the main and final glass explosions.",
"prompt_sha256": "b198e1ddb0d1d5c79cab984ce06e7944ecd4efbca41f105960c7312c540dd572",
"duration": 15,
"seed": 53,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "SHATTER.",
"zh": "SHATTER."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.027102902997285,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/12.mp4",
"poster": "assets/v2-15s/fasth3-4/12.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8980980,
"sha256": "a1ea7fe2db44a233bf603bdf078ffce957e1fb18c7b03a3866d3e2a2b041dbb6",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/12.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/12.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 13,
"slug": "shot-opens-tight-on-a-beautiful-woman-face-wind-452380",
"title": {
"en": "Cliffside Descent to the Ocean",
"zh": "从悬崖人像到俯拍海浪"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@umesh_ai",
"url": "https://x.com/umesh_ai/status/2082700637444452380"
},
"source_prompt": "The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below. Perform the described camera path as one continuous 15-second move, ending on the high overhead view of the lone figure and exploding surf.\n\noverall_soundscape: Strong coastal wind drives the coat fabric. Deep ocean roar becomes clearer as the camera descends, with violent waves striking rock beneath the final overhead view.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "8a9fb882646cdb521b55ea777eea72c255e9b3863f257602d34d5bda4da3b879",
"duration": 15,
"seed": 54,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Cliffside Descent to the Ocean",
"zh": "悬崖下行至海边"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 12.714291729032993,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/13.mp4",
"poster": "assets/v2-15s/fasth3-4/13.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 11083182,
"sha256": "1269e1e7391bd501e314964b88b1dae6be6681b7b2d3157db789d92666297742",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/13.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/13.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 14,
"slug": "ugc-style-skincare-video-featuring-a-realistic-young-wo-242986",
"title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "镜前护肤产品口播"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@ZaraIrahh",
"url": "https://x.com/ZaraIrahh/status/2083011066800242986"
},
"source_prompt": "UGC-style skincare video featuring a realistic young woman speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, vertical 9:16, social media ad quality, highly realistic natural skin texture, not overly retouched.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] UGC-style skincare video featuring a realistic young woman with a clear conversational English voice (S1) speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, landscape 16:9, social media ad quality, highly realistic natural skin texture, not overly retouched. Her short natural English presentation stays about her skincare routine. Use a continuous 15-second landscape smartphone shot framed to keep both her face and the bottle clearly visible. She presents the serum, applies it to her face, and returns her attention to the camera.\n\noverall_soundscape: Quiet bathroom room tone, small glass-bottle handling sounds and soft hand-on-skin movements accompany the presentation. Keep breaths and clothing movement natural and close.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "61f30fc7218593117f92f6c7e46d55e03de10b31642a4122a69a9eba43d40b4c",
"duration": 15,
"seed": 55,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "浴室镜前护肤 UGC 广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 12.73095204308629,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/14.mp4",
"poster": "assets/v2-15s/fasth3-4/14.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4560139,
"sha256": "edbbf90d2ddec9845281f4991701ba21544dee90c35df2a82fb359ba45f8d6a7",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/14.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/14.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 15,
"slug": "found-footage-film-with-335284",
"title": {
"en": "Horror Film Study 335284",
"zh": "渔夫接到诡异来电"
},
"category": "horror",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2084842607411335284"
},
"source_prompt": "\"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A distorted voice says, \"Finally. I’ve been trying to switch places with you.\" The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A low distorted English voice from the phone speaker (S1) says: [English] Finally. I’ve been trying to switch places with you. The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\" Keep the cheap boat-mounted camera viewpoint continuous for the 15-second scene, with only the specified white-flash glitch interrupting the image. Establish the fishing action, allow the phone line to finish, show the trapped image and then the wet double before the abrupt final black frame.\n\noverall_soundscape: Low lake water lapping, boat creaks and fishing-reel strain accompany the fisherman’s heavy breathing. Wet weeds scrape the hull, the phone glitches with clipped electronic noise, then water splashes as the soaked double enters the boat.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "778e3b3e9a6ca16d59312279338db624dd6281b72e5ade48402f63ed4b25f617",
"duration": 15,
"seed": 56,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Horror Film Study 335284",
"zh": "恐怖短片案例 335284"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.290331610944122,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/15.mp4",
"poster": "assets/v2-15s/fasth3-4/15.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9875469,
"sha256": "ed08b1518c4242806f80efac2b115a416ce7e4785ee1bc802b0c722d56b6849a",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/15.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/15.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 16,
"slug": "cinematic-sci-fi-suspense-video-a-realistic-spaceship-crew-in-024495",
"title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "太空船门后的小学教室"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085498501384024495"
},
"source_prompt": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic uniforms follows a mysterious distress signal and boards a dark, abandoned spacecraft drifting in space. Flashlights cut through darkness as they move cautiously through narrow metallic corridors. Warning lights flicker, metal creaks, and their boots echo on the floor. One crew member says, \"Signal's coming from deeper inside.\" They enter a large chamber filled with hundreds of empty cryogenic pods, all open and abandoned. Another crew member says nervously, \"Where did everyone go?\" They approach a final closed door at the far end. The captain slowly opens it, and instead of another ship compartment, it reveals a completely normal present-day elementary school classroom in bright daylight. Children sit at desks, backpacks on the floor, colorful posters on the walls, and a teacher at the front turns to the crew and scolds them: \"You're late again. Take your seats.\" The astronauts stand frozen in confusion at the doorway. Realistic tone, smooth camera movement, strong suspense buildup, with the final reveal clear, sudden, and absurd. Include realistic audio: radio static, low ship hum, footsteps, pod machinery ambience, door hiss, then ordinary classroom sounds and the teacher's voice.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic science-fiction suspense, realistic spaceship crew in full futuristic uniforms boarding a dark abandoned spacecraft in response to a distress signal. A smooth tracking shot follows them through a narrow metallic corridor; flashlight beams cross flickering warning lights. One crew member with a restrained mid-pitched English voice (S1) says: [English] Signal's coming from deeper inside. Radio static, a low ship hum, creaking metal and hard boot echoes accompany their advance.\n[Shot 2] At 00:04.500, the camera cuts to a wide view of a large chamber with hundreds of open, empty cryogenic pods. A second crew member with an uneasy, slightly higher English voice (S2) looks from the abandoned pods toward a final closed door and says nervously: [English] Where did everyone go? The captain approaches that door as the camera pushes forward with the crew. Pod machinery resonates beneath their footsteps.\n[Shot 3] At 00:09.000, the camera cuts behind the captain as the door hisses open. Bright daylight reveals an ordinary present-day elementary school classroom, children at desks, backpacks on the floor and colorful posters on the walls. A teacher at the front with a clear, matter-of-fact English voice (S3) turns toward the astronauts and scolds them: [English] You're late again. Take your seats. The astronauts remain frozen in confusion at the doorway. Hold the absurd reveal until the end while ordinary classroom room tone replaces the ship ambience.\n\noverall_soundscape: Radio static, a low ship hum, creaking metal, boot echoes and empty pod machinery fill the spacecraft. A door hiss gives way abruptly to ordinary classroom chair movement and quiet paper rustling.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "515b6f9c02a9aff5ebdd48c593434b27848f0a398afc1c0651103dfdd475c0da",
"duration": 15,
"seed": 57,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.769144061021507,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/16.mp4",
"poster": "assets/v2-15s/fasth3-4/16.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8488005,
"sha256": "453f8c0b801b032b06c1ff94d0b9e612cf141a7e2954f4edc59cd2f233e46854",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/16.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/16.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 17,
"slug": "skyship-flight-across-a-floating-kingdom-single-continu-344891",
"title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "穿越浮岛王国的飞艇"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@Strength04_X",
"url": "https://x.com/Strength04_X/status/2082712692159344891"
},
"source_prompt": "Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds. Complete the described path in one uninterrupted 15-second flight, establishing the skyship first and reserving the final seconds for the sunlight reveal beyond the storm.\n\noverall_soundscape: High-altitude wind and engine thrust deepen as the skyship enters thunder, lashing rain and cracking rock. Waterfalls rush past the hull before the storm sound falls behind the ship in open sunlight.\n\nnon_diegetic_music: Broad sustained strings and steady low orchestral percussion build through the storm, then open into a long brass-and-string chord in sunlight.",
"prompt_sha256": "15ed03a4ff9ad4d170616c8220d15f8c7ddadb16a059a522c567a1cede541838",
"duration": 15,
"seed": 58,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "飞空艇穿越浮空王国"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.923790429718792,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/17.mp4",
"poster": "assets/v2-15s/fasth3-4/17.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 13345460,
"sha256": "65ddfb7838bfc3f295707356b32649b7ba46196a62336e64cd15a72c7c717842",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/17.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/17.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 18,
"slug": "a-hyper-realistic-handheld-phone-video-of-a-quiet-suburban-495564",
"title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "后院上空突然坠落的天空"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2086879654116495564"
},
"source_prompt": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\" Use a single 15-second home-video take. Preserve the first several seconds of ordinary watering, begin the overhead crack after that peaceful lead-in, and end with the sky descending and camera falling.\n\noverall_soundscape: Ordinary backyard birds, distant traffic and a watering stream continue for several seconds. A sudden deep overhead crack becomes a heavy descending rumble, followed by startled breathing and the phone hitting the ground.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "42f8695e30223aeca06566bf0476f7882751c30d3db90ad0136585d517c6ba26",
"duration": 15,
"seed": 59,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.091396818868816,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/18.mp4",
"poster": "assets/v2-15s/fasth3-4/18.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 10860207,
"sha256": "a809574802cfe4457e63a4f60ae5f5152adc11354db92c0512c324e5c77d08a7",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/18.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/18.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 19,
"slug": "also-has-an-impressive-understanding-of-differen-911311",
"title": {
"en": "also has an impressive understanding of different cultures",
"zh": "穿行爪哇市场的肉丸摊贩"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@junwatu",
"url": "https://x.com/junwatu/status/2084840152715911311"
},
"source_prompt": "Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism. Nearby women bargaining in Javanese (S1) and vendors calling prices in Javanese (S2) remain overlapping background voices. Follow the vendor in one continuous 15-second handheld take, keeping the steaming bowl and his careful steps legible; finish with serving the bowl and the brief power flicker.\n\noverall_soundscape: Dense covered-market bustle includes footsteps, bowl clatter, steaming food and sacks brushing clothing. Slow motorcycle engines pass among the stalls, followed by a brief fluorescent power buzz and the uninterrupted market activity.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "644ee708e360b5247f1f86b9510bc54be46b1ede81be1dfd8e283227dfab466f",
"duration": 15,
"seed": 60,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "also has an impressive understanding of different cultures",
"zh": "also has an impressive understanding of different cultures"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 12.90002251509577,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/19.mp4",
"poster": "assets/v2-15s/fasth3-4/19.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 10891060,
"sha256": "3353dc19545d8ea0de2ec192396a91216d4df4afb358b2d538ebb2e3151ec293",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/19.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/19.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 405.68627888010815,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 20,
"slug": "16-9-cinematic-wuxia-mystery-set-in-a-bamboo-571041",
"title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "夜雪竹林里的武侠悬疑"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@AIwithAliya",
"url": "https://x.com/AIwithAliya/status/2083132770650571041"
},
"source_prompt": "A 16:9 cinematic wuxia mystery set in a bamboo forest at night. Use a low-saturation palette of cold blue, ink green, charcoal, and gray. Thin mist fills the forest and fine snow drifts through the air. The mood is austere, lethal, and controlled, with the tension of a martial-arts sect investigating a case and exchanging secret intelligence.\n\nDeep in the forest, dense vertical bamboo fills the background while cold white mist-light glows in the distance. Soft, out-of-focus leaves partially obscure the foreground, creating the sense of watching from within the grove. A cool, soft front-side key lights the actors’ faces; backlight keeps the foreground dark and the distance luminous. Use shallow depth of field so leaves, snow, and bamboo dissolve into soft bokeh.\nPrioritize facial close-ups and measured shot/reverse-shot coverage. Keep the rhythm restrained but tense. Photoreal period-drama production value, cinematic lighting, and no modern elements. No subtitles, on-screen text, watermarks, modern clothing or architecture, animation styling, over-smoothed skin, bright daylight, or comic performance.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic wuxia mystery in a bamboo forest at night, in a restrained cold-blue, ink-green, charcoal and gray palette. Thin mist and fine drifting snow surround two period-clothed martial-arts investigators. Dense vertical bamboo fills the background; soft foreground leaves partially obscure a tight facial close-up of the first investigator. Cool, soft front-side light reveals natural skin texture while cold white mist-light glows in the distance. The investigator looks toward the other person with measured, wary attention. No modern objects, text, subtitles, watermarks or comic performance.\n[Shot 2] At 00:05.000, the camera cuts to the other investigator in matching close-up for a restrained reverse angle. Keep the same cold lighting, shallow depth of field and softly blurred falling snow. A small shift of the eyes toward the dark grove conveys the exchange of secret intelligence without adding invented dialogue. Clothing rustles quietly under the bamboo wind.\n[Shot 3] At 00:10.000, the camera cuts back to the first investigator seen through out-of-focus leaves. A slow, small push-in follows a controlled glance into the luminous mist. Hold the austere, tense expression until the end, with the distant forest remaining still except for snow and moving leaves. Photoreal period-drama lighting, no animation styling, bright daylight or over-smoothed skin.\n\noverall_soundscape: Thin wind passes through bamboo leaves with occasional dry stalk creaks. Soft period-clothing movement and faint footsteps remain close while falling snow and distant forest space stay quiet.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "11628027721bd267f57680bb10ff37aff3623b6ea6bbfb064178144173e3a828",
"duration": 15,
"seed": 61,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "竹林武侠谜案"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.855335155967623,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/fasth3-4/20.mp4",
"poster": "assets/v2-15s/fasth3-4/20.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4361983,
"sha256": "f9e56708b01a7290fc25d84cf2aed2c80c8efce92b6fb836f5ccf62afcfd5c67",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/fasth3-4/20.mp4",
"dataset_path": "videos/v2-15s/fasth3-4/20.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 1,
"slug": "via-mlx-serve-by-561487",
"title": {
"en": "via MLX Serve by",
"zh": "小恐龙打喷嚏变出蝴蝶"
},
"category": "animation",
"source": {
"kind": "x",
"name": "@albertgao",
"url": "https://x.com/albertgao/status/2085137397134561487"
},
"source_prompt": "\"A cute little girl walks through a sunny forest when a small, friendly dinosaur steps out, waves, and says hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style.\"\n\n- time: 15min ish\n- 768x768",
"source_duration": "5s",
"prompt": "integrated_multimodal_description: [Shot 1] A cute little girl walks through a sunny forest when a small, friendly dinosaur with a light, cheerful English voice (S1) steps out, waves, and says: [English] hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style. Pace the meeting, sneeze and shared laughter over a continuous 15-second scene, allowing time to see the butterflies settle around both characters.\n\noverall_soundscape: Light forest wind, birds and small footsteps accompany the meeting. A comic sneeze releases a fluttering rush of butterfly wings, followed by the girl and dinosaur laughing together.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "954337465d95f9ab7b1fd512ba3879fa925f8e668f493e314c9d84a91e99fad3",
"duration": 15,
"seed": 42,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 5s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "via MLX Serve by",
"zh": "via MLX Serve by"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 31.44722088892013,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/01.mp4",
"poster": "assets/v2-15s/hyperflow-8/01.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3957369,
"sha256": "ec3d0204a2e0bc1d93061ca6d7635b2577babd6304e0ec69fef28f9ccbeb5c4f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/01.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/01.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 2,
"slug": "the-last-thing-you-see-in-your-first-and-last-290296",
"title": {
"en": "The last thing you see in your first and last",
"zh": "太空旅客窗外的黑洞"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@ivanfioravanti",
"url": "https://x.com/ivanfioravanti/status/2086553101029290296"
},
"source_prompt": "amateur handheld pov footage of a tourist in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: \"Wow look at that!\", then they show back the room",
"source_duration": "10s",
"prompt": "integrated_multimodal_description: [Shot 1] amateur handheld pov footage of a tourist with an excited, conversational English voice (S1) in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: [English] Wow look at that!, then they show back the room Use one uninterrupted 15-second handheld take. Switch off the room light around halfway through, hold the window view for the spoken reaction, then turn back to show the room.\n\noverall_soundscape: A steady enclosed-room ventilation hum and soft carpeted footsteps continue under slight handheld handling noise. A light-switch click marks the change to the darkened room.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "712f02d91f39b4d5a029cf4c3396609ac32d151511bf04949501f697139405e5",
"duration": 15,
"seed": 43,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 10s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "The last thing you see in your first and last",
"zh": "The last thing you see in your first and last"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 24.163682500831783,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/02.mp4",
"poster": "assets/v2-15s/hyperflow-8/02.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4973347,
"sha256": "29886ab08f788882031ddd88b26af0c705a34324f8308f921f76bece5ed6042c",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/02.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/02.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 3,
"slug": "ultra-cinematic-macro-shot-a-calm-mountain-lake-039920",
"title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "水滴汇成 MORNING"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2085622021812039920"
},
"source_prompt": "Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality. In one continuous 15-second camera move, establish the dawn lake first, raise and spiral the droplets in the middle, hold the completed readable word before its final liquid collapse.\n\noverall_soundscape: Quiet mountain wind, small lake ripples and faint distant dawn birds establish the lakeside. Delicate water movement grows into the rushing liquid-letter collapse and a crisp shower of droplets.\n\nnon_diegetic_music: Slow, sustained string notes and sparse high piano tones rise gently with the water lettering and resolve as the droplets fall.",
"prompt_sha256": "7abb51ba6a7327cb9266b13da9a8fa218928aeb0d97f444c82dc95ea21c245c2",
"duration": 15,
"seed": 44,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "Ultra cinematic macro shot: A calm mountain lake at dawn"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.701676580123603,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/03.mp4",
"poster": "assets/v2-15s/hyperflow-8/03.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4840961,
"sha256": "3326242d298f719888ba2075ef34d2b26721d5fa21b8b890f856c30a3735a9c6",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/03.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/03.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 4,
"slug": "strawberry-seasonal-match-cut-food-commercial-117943",
"title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓穿过四季落入奶油"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@HBCoop_",
"url": "https://x.com/HBCoop_/status/2082827829172117943"
},
"source_prompt": "Premium food commercial. Extreme macro shot of a single ripe strawberry falling toward a white ceramic bowl. Before it lands, the environment changes through four seamless seasonal match cuts: a sunlit spring greenhouse, a hot summer picnic, an autumn kitchen and a candlelit winter dining room. The strawberry stays in exactly the same screen position, size, orientation and downward motion through every transition. It lands in fresh cream during the winter scene, sending one graceful crown-shaped splash upward. Photorealistic food texture, high-speed product cinematography, clean commercial lighting. Each season has distinct environmental sound, but one continuous musical phrase connects the entire sequence. No hands, labels or additional fruit.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Photorealistic premium food commercial, extreme macro high-speed product cinematography. A single ripe strawberry falls toward a white ceramic bowl in a sunlit spring greenhouse. Its red seeds and moist surface are sharply resolved against soft green glasshouse bokeh. Keep the strawberry's screen position, size, orientation and downward motion consistent through every seasonal match cut. No hands, labels or additional fruit.\n[Shot 2] At 00:03.000, the shot transitions to a hot summer picnic in a seamless seasonal match cut around the same falling strawberry and bowl. Warm sunlight replaces the greenhouse light; distant insects accompany its uninterrupted descent.\n[Shot 3] At 00:06.000, the shot transitions to an autumn kitchen with amber window light, preserving the strawberry and bowl alignment and the same downward motion. Soft indoor room tone replaces the outdoor insects.\n[Shot 4] At 00:09.000, the shot transitions to a candlelit winter dining room. The strawberry completes its fall into fresh cream in the white bowl, sending one graceful crown-shaped splash upward. Hold the detailed splash and its settling ripples through the final seconds. One continuous musical phrase connects all four seasons.\n\noverall_soundscape: Spring birds and a gentle greenhouse breeze change to summer insects, quiet autumn kitchen ambience, then the still winter dining room. The final cream splash and falling droplets remain clearly audible.\n\nnon_diegetic_music: One uninterrupted light piano-and-plucked-string phrase runs at a moderate tempo across all four seasons, resolving with the strawberry splash.",
"prompt_sha256": "0c4614d6b1c88a86974b1393eab7a574817abc11c37bf72e3cb53e2d9bbf718f",
"duration": 15,
"seed": 45,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓四季匹配剪辑食品广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.846177545841783,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/04.mp4",
"poster": "assets/v2-15s/hyperflow-8/04.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1926025,
"sha256": "0e48a5923fb47456d97f0d7c8ede826a0847e77f590c38751e34469e1c23d0d9",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/04.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/04.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 5,
"slug": "stunning-action-packed-scene-following-a-little-robot-r-288619",
"title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "赛博城市里的机器人追逐"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@AIandDesign",
"url": "https://x.com/AIandDesign/status/2082522979339288619"
},
"source_prompt": "A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown. Maintain one continuous 15-second rear chase shot with brief arcs that reveal the pursuers when the robot looks back; preserve the robot’s identity and forward motion.\n\noverall_soundscape: Rapid metallic footfalls and small robot servo movements run through the city ambience. Heavy pursuing footsteps, drone rotors and passing machinery track the chase without intelligible dialogue.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be75f61112b0d84692fb5ce91ced32ae50c3f1a3a1ad6b115fba2089944d4073",
"duration": 15,
"seed": 46,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "小机器人赛博朋克逃亡"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 24.001835828181356,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/05.mp4",
"poster": "assets/v2-15s/hyperflow-8/05.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8365151,
"sha256": "5a5f5387a08887839fd19e863855fdad094eeea60f17b1c073bda1c73808f122",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/05.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/05.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 6,
"slug": "created-with-minimax-h3-max-120666",
"title": {
"en": "Created with MiniMax H3 Max",
"zh": "湖畔花束与风"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@TaliaAariz",
"url": "https://x.com/TaliaAariz/status/2095857062987120666"
},
"source_prompt": "A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field. Keep one continuous 15-second medium shot, with a very slow small push-in as she touches the flowers, smiles and finally closes her eyes to feel the breeze.\n\noverall_soundscape: Gentle shore wind moves the ribbons and reeds while small waves wash over the pebbles. Quiet bouquet and fabric rustles accompany her hand movements.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7e05bce3aa444d99e753af9bcc4865bd2934e2c9102f5a1ce63f952302516d15",
"duration": 15,
"seed": 47,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3 Max",
"zh": "Created with MiniMax H3 Max"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.892320454120636,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/06.mp4",
"poster": "assets/v2-15s/hyperflow-8/06.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3771535,
"sha256": "cf4bbb52bf1b4201bcc754a303d362d99a39385fd6c3fb0b44fb542bea1503e7",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/06.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/06.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 7,
"slug": "created-with-minimax-h3-on-989121",
"title": {
"en": "Created with MiniMax H3 on .",
"zh": "画家与湖畔城市天际线"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087408671101989121"
},
"source_prompt": "A young Western artist sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western artist with a relaxed conversational English voice (S1) sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience. Her brief natural English remarks follow her painting actions. Keep the artist, easel and skyline consistent over the 15-second take. Begin with her positioning the easel, follow several visible finishing brush strokes, then give the final canvas-to-skyline comparison time to read.\n\noverall_soundscape: Soft city-park wind, distant traffic and lakeside birds surround the easel. Light wood movements, brush-on-canvas friction and cloth handling follow the artist’s actions.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "f39be188be70b922d6337aa1436e276c98592d5adf01475408b43bbd227e6cbc",
"duration": 15,
"seed": 48,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Created with MiniMax H3 on .",
"zh": "Created with MiniMax H3 on ."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.84398057172075,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/07.mp4",
"poster": "assets/v2-15s/hyperflow-8/07.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4056216,
"sha256": "117ab831498a86b8358e40cebc9f29411d5128bbe55789ba448f97f8ed0a273f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/07.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/07.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 8,
"slug": "created-with-minimax-h3-522089",
"title": {
"en": "Created with MiniMax H3.",
"zh": "街头摄影师、老人和小狗"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087102541146522089"
},
"source_prompt": "A young Western female street photographer walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says “Look at that,” then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western female street photographer with a warm conversational English voice (S1) walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says: [English] Look at that, then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism. Over one continuous 15-second handheld shot, follow her noticing and photographing the café scene, then turn with her as she shows the camera display, speaks and continues walking.\n\noverall_soundscape: Layered downtown footsteps, distant traffic and soft café room spill continue beneath a camera shutter click. The dog shifts and lightly pants beside the seated man.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be91933cc24647e7105ff883517384c8c83c2ccef42801f72555f509c467e319",
"duration": 15,
"seed": 49,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3.",
"zh": "Created with MiniMax H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 24.031339339911938,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/08.mp4",
"poster": "assets/v2-15s/hyperflow-8/08.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 11448912,
"sha256": "5fb8dd1d15ec864526e9c57e5e799bf2fcd2c2f719473609d84b533754ddc97a",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/08.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/08.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 9,
"slug": "low-angle-fashion-tracking-film-062019",
"title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "小巷里的时装与歌声"
},
"category": "fashion",
"source": {
"kind": "x",
"name": "@Kiber_Alla",
"url": "https://x.com/Kiber_Alla/status/2083583963512062019"
},
"source_prompt": "Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. She sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. The woman, with a light melodic singing voice (S1), sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look. She performs a wordless vocal melody. Keep the low-angle backward tracking shot continuous for the full 15 seconds. Leave enough space for a full turn and return to eye contact without cutting.\n\noverall_soundscape: Light footsteps on stone, wind through the alley and natural fabric movement accompany the performance. Her soft laughter and breaths are synchronized to the visible movement.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7c2058cc5dc84a848b29169d61ecc607158e3ce4e68dd07d43622c10bfd34d91",
"duration": 15,
"seed": 50,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词或歌词；未编造逐字台词，歌唱使用无词旋律。"
],
"source_title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "低机位时尚跟拍短片"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.710000686813146,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/09.mp4",
"poster": "assets/v2-15s/hyperflow-8/09.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9387986,
"sha256": "b776b84381fd14e090c6f68370a596687be78c1a3ff8ed7980f75bd5c59ee001",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/09.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/09.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 10,
"slug": "generated-a-french-bulldog-with-minimax-h3-182507",
"title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "法国斗牛犬闯障碍"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085448660612182507"
},
"source_prompt": "A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose. Follow the complete obstacle sequence in one 15-second handheld take, with the bell contact visible in the final seconds.\n\noverall_soundscape: Backyard birds and distant neighborhood ambience surround quick paw taps and excited dog panting. Plastic poles rattle, the fabric tunnel scrapes, and a small bell rings once at the finish.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "b441f7d70cc0c7a54efccb096492e34839955225bc31781700cf560af438bc8d",
"duration": 15,
"seed": 51,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "Generated a french bulldog with MiniMax H3"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.865956634748727,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/10.mp4",
"poster": "assets/v2-15s/hyperflow-8/10.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 16751851,
"sha256": "088ee1cefac41673416633b823644e1235cece97d406c905f7c1da686648af22",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/10.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/10.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 11,
"slug": "created-with-hailuo-h3-182619",
"title": {
"en": "Created with Hailuo H3.",
"zh": "冰川崩塌组成 EXTINCTION"
},
"category": "title-sequence",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2090740724803182619"
},
"source_prompt": "Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX. Over 15 seconds, establish the glacier, follow the expanding fractures and airborne ice, hold the completed readable ice title, then finish on its violent shatter.\n\noverall_soundscape: Deep frozen-ocean wind surrounds spreading ice creaks and resonant fractures. Heavy ice collapse becomes a crystalline debris storm, followed by the sharp final title shatter.\n\nnon_diegetic_music: Sustained low strings swell with the ice fracture, joined by deep percussion at the shelf collapse and final title shatter.",
"prompt_sha256": "4450f406537c05f4eaf4263bdbb5e1e4d60f9b8e253dbe76aac8d0be5e2b24bd",
"duration": 15,
"seed": 52,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with Hailuo H3.",
"zh": "Created with Hailuo H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.703748255968094,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/11.mp4",
"poster": "assets/v2-15s/hyperflow-8/11.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6991570,
"sha256": "be32be7a2707097a2b5ef7e0147dbf2e25c5958f91a18aaf044c399192b37708",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/11.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/11.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 12,
"slug": "shatter-695817",
"title": {
"en": "SHATTER.",
"zh": "玻璃碎片组成 SHATTER"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2096554023809695817"
},
"source_prompt": "Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX. Keep the camera move continuous over 15 seconds: establish the floating monolith, enter the spreading fracture, orbit the suspended fragments, hold the completed title briefly, then finish on fragments expanding toward the lens.\n\noverall_soundscape: A projectile impact triggers fine traveling glass cracks and a swelling structural groan. Dense crystalline shattering and passing shards follow the orbit, with a second pressure-wave crack and outward burst at the end.\n\nnon_diegetic_music: A sparse low electronic pulse builds beneath bowed-metal tones, with two deep percussion hits at the main and final glass explosions.",
"prompt_sha256": "b198e1ddb0d1d5c79cab984ce06e7944ecd4efbca41f105960c7312c540dd572",
"duration": 15,
"seed": 53,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "SHATTER.",
"zh": "SHATTER."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.733664467930794,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/12.mp4",
"poster": "assets/v2-15s/hyperflow-8/12.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 7373635,
"sha256": "a062da968a7a60bb9edfb346bef1ac22f146742ec85805717618a9fb0cf9f7f8",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/12.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/12.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 13,
"slug": "shot-opens-tight-on-a-beautiful-woman-face-wind-452380",
"title": {
"en": "Cliffside Descent to the Ocean",
"zh": "从悬崖人像到俯拍海浪"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@umesh_ai",
"url": "https://x.com/umesh_ai/status/2082700637444452380"
},
"source_prompt": "The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below. Perform the described camera path as one continuous 15-second move, ending on the high overhead view of the lone figure and exploding surf.\n\noverall_soundscape: Strong coastal wind drives the coat fabric. Deep ocean roar becomes clearer as the camera descends, with violent waves striking rock beneath the final overhead view.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "8a9fb882646cdb521b55ea777eea72c255e9b3863f257602d34d5bda4da3b879",
"duration": 15,
"seed": 54,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Cliffside Descent to the Ocean",
"zh": "悬崖下行至海边"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.51624410180375,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/13.mp4",
"poster": "assets/v2-15s/hyperflow-8/13.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8705354,
"sha256": "30147ab7e53a46f154887e2dc9088eaa2cee82ebaad202295d5ca01aed313f6f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/13.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/13.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 14,
"slug": "ugc-style-skincare-video-featuring-a-realistic-young-wo-242986",
"title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "镜前护肤产品口播"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@ZaraIrahh",
"url": "https://x.com/ZaraIrahh/status/2083011066800242986"
},
"source_prompt": "UGC-style skincare video featuring a realistic young woman speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, vertical 9:16, social media ad quality, highly realistic natural skin texture, not overly retouched.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] UGC-style skincare video featuring a realistic young woman with a clear conversational English voice (S1) speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, landscape 16:9, social media ad quality, highly realistic natural skin texture, not overly retouched. Her short natural English presentation stays about her skincare routine. Use a continuous 15-second landscape smartphone shot framed to keep both her face and the bottle clearly visible. She presents the serum, applies it to her face, and returns her attention to the camera.\n\noverall_soundscape: Quiet bathroom room tone, small glass-bottle handling sounds and soft hand-on-skin movements accompany the presentation. Keep breaths and clothing movement natural and close.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "61f30fc7218593117f92f6c7e46d55e03de10b31642a4122a69a9eba43d40b4c",
"duration": 15,
"seed": 55,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "浴室镜前护肤 UGC 广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.412956648040563,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/14.mp4",
"poster": "assets/v2-15s/hyperflow-8/14.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3956592,
"sha256": "c6efbe00f55c5e7736ebcc6ac56f5a4e4603300f667528537d64c3e047b1f1f6",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/14.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/14.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 15,
"slug": "found-footage-film-with-335284",
"title": {
"en": "Horror Film Study 335284",
"zh": "渔夫接到诡异来电"
},
"category": "horror",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2084842607411335284"
},
"source_prompt": "\"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A distorted voice says, \"Finally. I’ve been trying to switch places with you.\" The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A low distorted English voice from the phone speaker (S1) says: [English] Finally. I’ve been trying to switch places with you. The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\" Keep the cheap boat-mounted camera viewpoint continuous for the 15-second scene, with only the specified white-flash glitch interrupting the image. Establish the fishing action, allow the phone line to finish, show the trapped image and then the wet double before the abrupt final black frame.\n\noverall_soundscape: Low lake water lapping, boat creaks and fishing-reel strain accompany the fisherman’s heavy breathing. Wet weeds scrape the hull, the phone glitches with clipped electronic noise, then water splashes as the soaked double enters the boat.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "778e3b3e9a6ca16d59312279338db624dd6281b72e5ade48402f63ed4b25f617",
"duration": 15,
"seed": 56,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Horror Film Study 335284",
"zh": "恐怖短片案例 335284"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.726625872775912,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/15.mp4",
"poster": "assets/v2-15s/hyperflow-8/15.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5922903,
"sha256": "4de8b4c7568ff123898736c87dcd8b8f9454877b6ef358b8db58be8a091089ce",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/15.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/15.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 16,
"slug": "cinematic-sci-fi-suspense-video-a-realistic-spaceship-crew-in-024495",
"title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "太空船门后的小学教室"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085498501384024495"
},
"source_prompt": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic uniforms follows a mysterious distress signal and boards a dark, abandoned spacecraft drifting in space. Flashlights cut through darkness as they move cautiously through narrow metallic corridors. Warning lights flicker, metal creaks, and their boots echo on the floor. One crew member says, \"Signal's coming from deeper inside.\" They enter a large chamber filled with hundreds of empty cryogenic pods, all open and abandoned. Another crew member says nervously, \"Where did everyone go?\" They approach a final closed door at the far end. The captain slowly opens it, and instead of another ship compartment, it reveals a completely normal present-day elementary school classroom in bright daylight. Children sit at desks, backpacks on the floor, colorful posters on the walls, and a teacher at the front turns to the crew and scolds them: \"You're late again. Take your seats.\" The astronauts stand frozen in confusion at the doorway. Realistic tone, smooth camera movement, strong suspense buildup, with the final reveal clear, sudden, and absurd. Include realistic audio: radio static, low ship hum, footsteps, pod machinery ambience, door hiss, then ordinary classroom sounds and the teacher's voice.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic science-fiction suspense, realistic spaceship crew in full futuristic uniforms boarding a dark abandoned spacecraft in response to a distress signal. A smooth tracking shot follows them through a narrow metallic corridor; flashlight beams cross flickering warning lights. One crew member with a restrained mid-pitched English voice (S1) says: [English] Signal's coming from deeper inside. Radio static, a low ship hum, creaking metal and hard boot echoes accompany their advance.\n[Shot 2] At 00:04.500, the camera cuts to a wide view of a large chamber with hundreds of open, empty cryogenic pods. A second crew member with an uneasy, slightly higher English voice (S2) looks from the abandoned pods toward a final closed door and says nervously: [English] Where did everyone go? The captain approaches that door as the camera pushes forward with the crew. Pod machinery resonates beneath their footsteps.\n[Shot 3] At 00:09.000, the camera cuts behind the captain as the door hisses open. Bright daylight reveals an ordinary present-day elementary school classroom, children at desks, backpacks on the floor and colorful posters on the walls. A teacher at the front with a clear, matter-of-fact English voice (S3) turns toward the astronauts and scolds them: [English] You're late again. Take your seats. The astronauts remain frozen in confusion at the doorway. Hold the absurd reveal until the end while ordinary classroom room tone replaces the ship ambience.\n\noverall_soundscape: Radio static, a low ship hum, creaking metal, boot echoes and empty pod machinery fill the spacecraft. A door hiss gives way abruptly to ordinary classroom chair movement and quiet paper rustling.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "515b6f9c02a9aff5ebdd48c593434b27848f0a398afc1c0651103dfdd475c0da",
"duration": 15,
"seed": 57,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.7857776992023,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/16.mp4",
"poster": "assets/v2-15s/hyperflow-8/16.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5037351,
"sha256": "1c859affd1d6a1fbdba93bf8b545553ccefabcb063b2a6b3015742d037b6eb89",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/16.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/16.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 17,
"slug": "skyship-flight-across-a-floating-kingdom-single-continu-344891",
"title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "穿越浮岛王国的飞艇"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@Strength04_X",
"url": "https://x.com/Strength04_X/status/2082712692159344891"
},
"source_prompt": "Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds. Complete the described path in one uninterrupted 15-second flight, establishing the skyship first and reserving the final seconds for the sunlight reveal beyond the storm.\n\noverall_soundscape: High-altitude wind and engine thrust deepen as the skyship enters thunder, lashing rain and cracking rock. Waterfalls rush past the hull before the storm sound falls behind the ship in open sunlight.\n\nnon_diegetic_music: Broad sustained strings and steady low orchestral percussion build through the storm, then open into a long brass-and-string chord in sunlight.",
"prompt_sha256": "15ed03a4ff9ad4d170616c8220d15f8c7ddadb16a059a522c567a1cede541838",
"duration": 15,
"seed": 58,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "飞空艇穿越浮空王国"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.475021732971072,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/17.mp4",
"poster": "assets/v2-15s/hyperflow-8/17.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 7945892,
"sha256": "b88420a0c61557f419b201fdf4d705aea052bcd62ce4650db2823e0d738c0265",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/17.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/17.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 18,
"slug": "a-hyper-realistic-handheld-phone-video-of-a-quiet-suburban-495564",
"title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "后院上空突然坠落的天空"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2086879654116495564"
},
"source_prompt": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\" Use a single 15-second home-video take. Preserve the first several seconds of ordinary watering, begin the overhead crack after that peaceful lead-in, and end with the sky descending and camera falling.\n\noverall_soundscape: Ordinary backyard birds, distant traffic and a watering stream continue for several seconds. A sudden deep overhead crack becomes a heavy descending rumble, followed by startled breathing and the phone hitting the ground.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "42f8695e30223aeca06566bf0476f7882751c30d3db90ad0136585d517c6ba26",
"duration": 15,
"seed": 59,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 24.25950285512954,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/18.mp4",
"poster": "assets/v2-15s/hyperflow-8/18.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9023305,
"sha256": "eb7a6df9d2979f591d1850c4697b5f7c1a645824a67252e5ba831859e2df1bd0",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/18.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/18.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 19,
"slug": "also-has-an-impressive-understanding-of-differen-911311",
"title": {
"en": "also has an impressive understanding of different cultures",
"zh": "穿行爪哇市场的肉丸摊贩"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@junwatu",
"url": "https://x.com/junwatu/status/2084840152715911311"
},
"source_prompt": "Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism. Nearby women bargaining in Javanese (S1) and vendors calling prices in Javanese (S2) remain overlapping background voices. Follow the vendor in one continuous 15-second handheld take, keeping the steaming bowl and his careful steps legible; finish with serving the bowl and the brief power flicker.\n\noverall_soundscape: Dense covered-market bustle includes footsteps, bowl clatter, steaming food and sacks brushing clothing. Slow motorcycle engines pass among the stalls, followed by a brief fluorescent power buzz and the uninterrupted market activity.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "644ee708e360b5247f1f86b9510bc54be46b1ede81be1dfd8e283227dfab466f",
"duration": 15,
"seed": 60,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "also has an impressive understanding of different cultures",
"zh": "also has an impressive understanding of different cultures"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.886915264185518,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/19.mp4",
"poster": "assets/v2-15s/hyperflow-8/19.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 7437662,
"sha256": "71ebf0e6394afd8de29ada1beaccd0c2f8f1c228abd4c576f99e38855971c08f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/19.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/19.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 424.0454792641103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 20,
"slug": "16-9-cinematic-wuxia-mystery-set-in-a-bamboo-571041",
"title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "夜雪竹林里的武侠悬疑"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@AIwithAliya",
"url": "https://x.com/AIwithAliya/status/2083132770650571041"
},
"source_prompt": "A 16:9 cinematic wuxia mystery set in a bamboo forest at night. Use a low-saturation palette of cold blue, ink green, charcoal, and gray. Thin mist fills the forest and fine snow drifts through the air. The mood is austere, lethal, and controlled, with the tension of a martial-arts sect investigating a case and exchanging secret intelligence.\n\nDeep in the forest, dense vertical bamboo fills the background while cold white mist-light glows in the distance. Soft, out-of-focus leaves partially obscure the foreground, creating the sense of watching from within the grove. A cool, soft front-side key lights the actors’ faces; backlight keeps the foreground dark and the distance luminous. Use shallow depth of field so leaves, snow, and bamboo dissolve into soft bokeh.\nPrioritize facial close-ups and measured shot/reverse-shot coverage. Keep the rhythm restrained but tense. Photoreal period-drama production value, cinematic lighting, and no modern elements. No subtitles, on-screen text, watermarks, modern clothing or architecture, animation styling, over-smoothed skin, bright daylight, or comic performance.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic wuxia mystery in a bamboo forest at night, in a restrained cold-blue, ink-green, charcoal and gray palette. Thin mist and fine drifting snow surround two period-clothed martial-arts investigators. Dense vertical bamboo fills the background; soft foreground leaves partially obscure a tight facial close-up of the first investigator. Cool, soft front-side light reveals natural skin texture while cold white mist-light glows in the distance. The investigator looks toward the other person with measured, wary attention. No modern objects, text, subtitles, watermarks or comic performance.\n[Shot 2] At 00:05.000, the camera cuts to the other investigator in matching close-up for a restrained reverse angle. Keep the same cold lighting, shallow depth of field and softly blurred falling snow. A small shift of the eyes toward the dark grove conveys the exchange of secret intelligence without adding invented dialogue. Clothing rustles quietly under the bamboo wind.\n[Shot 3] At 00:10.000, the camera cuts back to the first investigator seen through out-of-focus leaves. A slow, small push-in follows a controlled glance into the luminous mist. Hold the austere, tense expression until the end, with the distant forest remaining still except for snow and moving leaves. Photoreal period-drama lighting, no animation styling, bright daylight or over-smoothed skin.\n\noverall_soundscape: Thin wind passes through bamboo leaves with occasional dry stalk creaks. Soft period-clothing movement and faint footsteps remain close while falling snow and distant forest space stay quiet.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "11628027721bd267f57680bb10ff37aff3623b6ea6bbfb064178144173e3a828",
"duration": 15,
"seed": 61,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "竹林武侠谜案"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 23.756963307969272,
"observed_sparse_calls": 288,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/hyperflow-8/20.mp4",
"poster": "assets/v2-15s/hyperflow-8/20.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 2224088,
"sha256": "da0a26f5047f011dea91c347b8300709f458951520b92f8142f96634e3764226",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/hyperflow-8/20.mp4",
"dataset_path": "videos/v2-15s/hyperflow-8/20.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 1,
"slug": "via-mlx-serve-by-561487",
"title": {
"en": "via MLX Serve by",
"zh": "小恐龙打喷嚏变出蝴蝶"
},
"category": "animation",
"source": {
"kind": "x",
"name": "@albertgao",
"url": "https://x.com/albertgao/status/2085137397134561487"
},
"source_prompt": "\"A cute little girl walks through a sunny forest when a small, friendly dinosaur steps out, waves, and says hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style.\"\n\n- time: 15min ish\n- 768x768",
"source_duration": "5s",
"prompt": "integrated_multimodal_description: [Shot 1] A cute little girl walks through a sunny forest when a small, friendly dinosaur with a light, cheerful English voice (S1) steps out, waves, and says: [English] hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style. Pace the meeting, sneeze and shared laughter over a continuous 15-second scene, allowing time to see the butterflies settle around both characters.\n\noverall_soundscape: Light forest wind, birds and small footsteps accompany the meeting. A comic sneeze releases a fluttering rush of butterfly wings, followed by the girl and dinosaur laughing together.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "954337465d95f9ab7b1fd512ba3879fa925f8e668f493e314c9d84a91e99fad3",
"duration": 15,
"seed": 42,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 5s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "via MLX Serve by",
"zh": "via MLX Serve by"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 20.255016945768148,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/01.mp4",
"poster": "assets/v2-15s/lightx2v-4/01.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5142008,
"sha256": "4da69e6aefa643d09e4f1dc828d000595704bf47c8c9405702280d7eee435c2b",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/01.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/01.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 2,
"slug": "the-last-thing-you-see-in-your-first-and-last-290296",
"title": {
"en": "The last thing you see in your first and last",
"zh": "太空旅客窗外的黑洞"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@ivanfioravanti",
"url": "https://x.com/ivanfioravanti/status/2086553101029290296"
},
"source_prompt": "amateur handheld pov footage of a tourist in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: \"Wow look at that!\", then they show back the room",
"source_duration": "10s",
"prompt": "integrated_multimodal_description: [Shot 1] amateur handheld pov footage of a tourist with an excited, conversational English voice (S1) in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: [English] Wow look at that!, then they show back the room Use one uninterrupted 15-second handheld take. Switch off the room light around halfway through, hold the window view for the spoken reaction, then turn back to show the room.\n\noverall_soundscape: A steady enclosed-room ventilation hum and soft carpeted footsteps continue under slight handheld handling noise. A light-switch click marks the change to the darkened room.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "712f02d91f39b4d5a029cf4c3396609ac32d151511bf04949501f697139405e5",
"duration": 15,
"seed": 43,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 10s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "The last thing you see in your first and last",
"zh": "The last thing you see in your first and last"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 13.716562563087791,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/02.mp4",
"poster": "assets/v2-15s/lightx2v-4/02.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6542142,
"sha256": "581de6abe2dfddc1e4e6ddaaa79866d878d39805279caa989bab1866e72c9cbe",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/02.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/02.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 3,
"slug": "ultra-cinematic-macro-shot-a-calm-mountain-lake-039920",
"title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "水滴汇成 MORNING"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2085622021812039920"
},
"source_prompt": "Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality. In one continuous 15-second camera move, establish the dawn lake first, raise and spiral the droplets in the middle, hold the completed readable word before its final liquid collapse.\n\noverall_soundscape: Quiet mountain wind, small lake ripples and faint distant dawn birds establish the lakeside. Delicate water movement grows into the rushing liquid-letter collapse and a crisp shower of droplets.\n\nnon_diegetic_music: Slow, sustained string notes and sparse high piano tones rise gently with the water lettering and resolve as the droplets fall.",
"prompt_sha256": "7abb51ba6a7327cb9266b13da9a8fa218928aeb0d97f444c82dc95ea21c245c2",
"duration": 15,
"seed": 44,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "Ultra cinematic macro shot: A calm mountain lake at dawn"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 18.244925754144788,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/03.mp4",
"poster": "assets/v2-15s/lightx2v-4/03.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6477081,
"sha256": "e8c26039001c84bd402131b8f61cf482862bc84477e481015dcbb8f674203fba",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/03.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/03.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 4,
"slug": "strawberry-seasonal-match-cut-food-commercial-117943",
"title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓穿过四季落入奶油"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@HBCoop_",
"url": "https://x.com/HBCoop_/status/2082827829172117943"
},
"source_prompt": "Premium food commercial. Extreme macro shot of a single ripe strawberry falling toward a white ceramic bowl. Before it lands, the environment changes through four seamless seasonal match cuts: a sunlit spring greenhouse, a hot summer picnic, an autumn kitchen and a candlelit winter dining room. The strawberry stays in exactly the same screen position, size, orientation and downward motion through every transition. It lands in fresh cream during the winter scene, sending one graceful crown-shaped splash upward. Photorealistic food texture, high-speed product cinematography, clean commercial lighting. Each season has distinct environmental sound, but one continuous musical phrase connects the entire sequence. No hands, labels or additional fruit.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Photorealistic premium food commercial, extreme macro high-speed product cinematography. A single ripe strawberry falls toward a white ceramic bowl in a sunlit spring greenhouse. Its red seeds and moist surface are sharply resolved against soft green glasshouse bokeh. Keep the strawberry's screen position, size, orientation and downward motion consistent through every seasonal match cut. No hands, labels or additional fruit.\n[Shot 2] At 00:03.000, the shot transitions to a hot summer picnic in a seamless seasonal match cut around the same falling strawberry and bowl. Warm sunlight replaces the greenhouse light; distant insects accompany its uninterrupted descent.\n[Shot 3] At 00:06.000, the shot transitions to an autumn kitchen with amber window light, preserving the strawberry and bowl alignment and the same downward motion. Soft indoor room tone replaces the outdoor insects.\n[Shot 4] At 00:09.000, the shot transitions to a candlelit winter dining room. The strawberry completes its fall into fresh cream in the white bowl, sending one graceful crown-shaped splash upward. Hold the detailed splash and its settling ripples through the final seconds. One continuous musical phrase connects all four seasons.\n\noverall_soundscape: Spring birds and a gentle greenhouse breeze change to summer insects, quiet autumn kitchen ambience, then the still winter dining room. The final cream splash and falling droplets remain clearly audible.\n\nnon_diegetic_music: One uninterrupted light piano-and-plucked-string phrase runs at a moderate tempo across all four seasons, resolving with the strawberry splash.",
"prompt_sha256": "0c4614d6b1c88a86974b1393eab7a574817abc11c37bf72e3cb53e2d9bbf718f",
"duration": 15,
"seed": 45,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓四季匹配剪辑食品广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 16.116741803009063,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/04.mp4",
"poster": "assets/v2-15s/lightx2v-4/04.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 2238263,
"sha256": "2feb84ee84b23fd71ba578bcd4b6bc098a083f400b0810bc3370d61c04ff5116",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/04.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/04.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 5,
"slug": "stunning-action-packed-scene-following-a-little-robot-r-288619",
"title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "赛博城市里的机器人追逐"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@AIandDesign",
"url": "https://x.com/AIandDesign/status/2082522979339288619"
},
"source_prompt": "A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown. Maintain one continuous 15-second rear chase shot with brief arcs that reveal the pursuers when the robot looks back; preserve the robot’s identity and forward motion.\n\noverall_soundscape: Rapid metallic footfalls and small robot servo movements run through the city ambience. Heavy pursuing footsteps, drone rotors and passing machinery track the chase without intelligible dialogue.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be75f61112b0d84692fb5ce91ced32ae50c3f1a3a1ad6b115fba2089944d4073",
"duration": 15,
"seed": 46,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "小机器人赛博朋克逃亡"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.297106197103858,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/05.mp4",
"poster": "assets/v2-15s/lightx2v-4/05.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 10620006,
"sha256": "8f16f00bbb3e6ebb7d9f7f704272365dddff319274fd9d826d68488cd43e6a18",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/05.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/05.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 6,
"slug": "created-with-minimax-h3-max-120666",
"title": {
"en": "Created with MiniMax H3 Max",
"zh": "湖畔花束与风"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@TaliaAariz",
"url": "https://x.com/TaliaAariz/status/2095857062987120666"
},
"source_prompt": "A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field. Keep one continuous 15-second medium shot, with a very slow small push-in as she touches the flowers, smiles and finally closes her eyes to feel the breeze.\n\noverall_soundscape: Gentle shore wind moves the ribbons and reeds while small waves wash over the pebbles. Quiet bouquet and fabric rustles accompany her hand movements.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7e05bce3aa444d99e753af9bcc4865bd2934e2c9102f5a1ce63f952302516d15",
"duration": 15,
"seed": 47,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3 Max",
"zh": "Created with MiniMax H3 Max"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 20.539152700453997,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/06.mp4",
"poster": "assets/v2-15s/lightx2v-4/06.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4285643,
"sha256": "a5a4d1c83151ed98f369117c5fb540b6bab7e94e4b312cf8cb2435e607437d3b",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/06.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/06.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 7,
"slug": "created-with-minimax-h3-on-989121",
"title": {
"en": "Created with MiniMax H3 on .",
"zh": "画家与湖畔城市天际线"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087408671101989121"
},
"source_prompt": "A young Western artist sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western artist with a relaxed conversational English voice (S1) sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience. Her brief natural English remarks follow her painting actions. Keep the artist, easel and skyline consistent over the 15-second take. Begin with her positioning the easel, follow several visible finishing brush strokes, then give the final canvas-to-skyline comparison time to read.\n\noverall_soundscape: Soft city-park wind, distant traffic and lakeside birds surround the easel. Light wood movements, brush-on-canvas friction and cloth handling follow the artist’s actions.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "f39be188be70b922d6337aa1436e276c98592d5adf01475408b43bbd227e6cbc",
"duration": 15,
"seed": 48,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Created with MiniMax H3 on .",
"zh": "Created with MiniMax H3 on ."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 17.762425834778696,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/07.mp4",
"poster": "assets/v2-15s/lightx2v-4/07.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3574357,
"sha256": "5d0518131fee9833d813d3a67e140ef81b3223837410bfc8d9cd27cbb6f799f1",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/07.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/07.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 8,
"slug": "created-with-minimax-h3-522089",
"title": {
"en": "Created with MiniMax H3.",
"zh": "街头摄影师、老人和小狗"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087102541146522089"
},
"source_prompt": "A young Western female street photographer walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says “Look at that,” then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western female street photographer with a warm conversational English voice (S1) walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says: [English] Look at that, then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism. Over one continuous 15-second handheld shot, follow her noticing and photographing the café scene, then turn with her as she shows the camera display, speaks and continues walking.\n\noverall_soundscape: Layered downtown footsteps, distant traffic and soft café room spill continue beneath a camera shutter click. The dog shifts and lightly pants beside the seated man.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be91933cc24647e7105ff883517384c8c83c2ccef42801f72555f509c467e319",
"duration": 15,
"seed": 49,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3.",
"zh": "Created with MiniMax H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.008996727876365,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/08.mp4",
"poster": "assets/v2-15s/lightx2v-4/08.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 13065707,
"sha256": "9765bf4815257fb4c0160762b9d542f0d6c0b4ae33d8d855a6764563e4a1336f",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/08.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/08.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 9,
"slug": "low-angle-fashion-tracking-film-062019",
"title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "小巷里的时装与歌声"
},
"category": "fashion",
"source": {
"kind": "x",
"name": "@Kiber_Alla",
"url": "https://x.com/Kiber_Alla/status/2083583963512062019"
},
"source_prompt": "Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. She sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. The woman, with a light melodic singing voice (S1), sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look. She performs a wordless vocal melody. Keep the low-angle backward tracking shot continuous for the full 15 seconds. Leave enough space for a full turn and return to eye contact without cutting.\n\noverall_soundscape: Light footsteps on stone, wind through the alley and natural fabric movement accompany the performance. Her soft laughter and breaths are synchronized to the visible movement.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7c2058cc5dc84a848b29169d61ecc607158e3ce4e68dd07d43622c10bfd34d91",
"duration": 15,
"seed": 50,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词或歌词；未编造逐字台词，歌唱使用无词旋律。"
],
"source_title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "低机位时尚跟拍短片"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 16.49164312472567,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/09.mp4",
"poster": "assets/v2-15s/lightx2v-4/09.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9378621,
"sha256": "678760472fde5af4fca5d0ec8b3bcae07923cb356f22dd082feb0b44a463d410",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/09.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/09.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 10,
"slug": "generated-a-french-bulldog-with-minimax-h3-182507",
"title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "法国斗牛犬闯障碍"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085448660612182507"
},
"source_prompt": "A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose. Follow the complete obstacle sequence in one 15-second handheld take, with the bell contact visible in the final seconds.\n\noverall_soundscape: Backyard birds and distant neighborhood ambience surround quick paw taps and excited dog panting. Plastic poles rattle, the fabric tunnel scrapes, and a small bell rings once at the finish.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "b441f7d70cc0c7a54efccb096492e34839955225bc31781700cf560af438bc8d",
"duration": 15,
"seed": 51,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "Generated a french bulldog with MiniMax H3"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.970872349105775,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/10.mp4",
"poster": "assets/v2-15s/lightx2v-4/10.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 30330339,
"sha256": "2973b567354c664fa0787a4f6fc7839664cd48a2a53e5f8c964cf71846a261f4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/10.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/10.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 11,
"slug": "created-with-hailuo-h3-182619",
"title": {
"en": "Created with Hailuo H3.",
"zh": "冰川崩塌组成 EXTINCTION"
},
"category": "title-sequence",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2090740724803182619"
},
"source_prompt": "Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX. Over 15 seconds, establish the glacier, follow the expanding fractures and airborne ice, hold the completed readable ice title, then finish on its violent shatter.\n\noverall_soundscape: Deep frozen-ocean wind surrounds spreading ice creaks and resonant fractures. Heavy ice collapse becomes a crystalline debris storm, followed by the sharp final title shatter.\n\nnon_diegetic_music: Sustained low strings swell with the ice fracture, joined by deep percussion at the shelf collapse and final title shatter.",
"prompt_sha256": "4450f406537c05f4eaf4263bdbb5e1e4d60f9b8e253dbe76aac8d0be5e2b24bd",
"duration": 15,
"seed": 52,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with Hailuo H3.",
"zh": "Created with Hailuo H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.799380616284907,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/11.mp4",
"poster": "assets/v2-15s/lightx2v-4/11.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8639754,
"sha256": "d5c4831cc35b069ed7127693276f1e630f1d3fcc162d6709f93d35839f494ba3",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/11.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/11.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 12,
"slug": "shatter-695817",
"title": {
"en": "SHATTER.",
"zh": "玻璃碎片组成 SHATTER"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2096554023809695817"
},
"source_prompt": "Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX. Keep the camera move continuous over 15 seconds: establish the floating monolith, enter the spreading fracture, orbit the suspended fragments, hold the completed title briefly, then finish on fragments expanding toward the lens.\n\noverall_soundscape: A projectile impact triggers fine traveling glass cracks and a swelling structural groan. Dense crystalline shattering and passing shards follow the orbit, with a second pressure-wave crack and outward burst at the end.\n\nnon_diegetic_music: A sparse low electronic pulse builds beneath bowed-metal tones, with two deep percussion hits at the main and final glass explosions.",
"prompt_sha256": "b198e1ddb0d1d5c79cab984ce06e7944ecd4efbca41f105960c7312c540dd572",
"duration": 15,
"seed": 53,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "SHATTER.",
"zh": "SHATTER."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.225255922880024,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/12.mp4",
"poster": "assets/v2-15s/lightx2v-4/12.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9688152,
"sha256": "79d2e7b5fd8cb7a542ae05d2bd8a6eee266dc487ccf6e1df09f2ff95de502ecb",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/12.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/12.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 13,
"slug": "shot-opens-tight-on-a-beautiful-woman-face-wind-452380",
"title": {
"en": "Cliffside Descent to the Ocean",
"zh": "从悬崖人像到俯拍海浪"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@umesh_ai",
"url": "https://x.com/umesh_ai/status/2082700637444452380"
},
"source_prompt": "The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below. Perform the described camera path as one continuous 15-second move, ending on the high overhead view of the lone figure and exploding surf.\n\noverall_soundscape: Strong coastal wind drives the coat fabric. Deep ocean roar becomes clearer as the camera descends, with violent waves striking rock beneath the final overhead view.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "8a9fb882646cdb521b55ea777eea72c255e9b3863f257602d34d5bda4da3b879",
"duration": 15,
"seed": 54,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Cliffside Descent to the Ocean",
"zh": "悬崖下行至海边"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 12.783019708935171,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/13.mp4",
"poster": "assets/v2-15s/lightx2v-4/13.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 11382393,
"sha256": "8f04be08d61fb0eeac292a1e50db564cb6a5c4f6a533e2532e9cc05fe2e3173c",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/13.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/13.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 14,
"slug": "ugc-style-skincare-video-featuring-a-realistic-young-wo-242986",
"title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "镜前护肤产品口播"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@ZaraIrahh",
"url": "https://x.com/ZaraIrahh/status/2083011066800242986"
},
"source_prompt": "UGC-style skincare video featuring a realistic young woman speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, vertical 9:16, social media ad quality, highly realistic natural skin texture, not overly retouched.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] UGC-style skincare video featuring a realistic young woman with a clear conversational English voice (S1) speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, landscape 16:9, social media ad quality, highly realistic natural skin texture, not overly retouched. Her short natural English presentation stays about her skincare routine. Use a continuous 15-second landscape smartphone shot framed to keep both her face and the bottle clearly visible. She presents the serum, applies it to her face, and returns her attention to the camera.\n\noverall_soundscape: Quiet bathroom room tone, small glass-bottle handling sounds and soft hand-on-skin movements accompany the presentation. Keep breaths and clothing movement natural and close.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "61f30fc7218593117f92f6c7e46d55e03de10b31642a4122a69a9eba43d40b4c",
"duration": 15,
"seed": 55,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "浴室镜前护肤 UGC 广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 12.813158273231238,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/14.mp4",
"poster": "assets/v2-15s/lightx2v-4/14.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3760929,
"sha256": "f12345b2b2879ccf185824d6005ba44e9441d41c5ddf2497384b58d98c2ae188",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/14.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/14.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 15,
"slug": "found-footage-film-with-335284",
"title": {
"en": "Horror Film Study 335284",
"zh": "渔夫接到诡异来电"
},
"category": "horror",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2084842607411335284"
},
"source_prompt": "\"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A distorted voice says, \"Finally. I’ve been trying to switch places with you.\" The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A low distorted English voice from the phone speaker (S1) says: [English] Finally. I’ve been trying to switch places with you. The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\" Keep the cheap boat-mounted camera viewpoint continuous for the 15-second scene, with only the specified white-flash glitch interrupting the image. Establish the fishing action, allow the phone line to finish, show the trapped image and then the wet double before the abrupt final black frame.\n\noverall_soundscape: Low lake water lapping, boat creaks and fishing-reel strain accompany the fisherman’s heavy breathing. Wet weeds scrape the hull, the phone glitches with clipped electronic noise, then water splashes as the soaked double enters the boat.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "778e3b3e9a6ca16d59312279338db624dd6281b72e5ade48402f63ed4b25f617",
"duration": 15,
"seed": 56,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Horror Film Study 335284",
"zh": "恐怖短片案例 335284"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.542793967295438,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/15.mp4",
"poster": "assets/v2-15s/lightx2v-4/15.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 7156287,
"sha256": "49106d72a9706c94dc771bfbcd63f14177fcd0e1b05b8deb47ac1f5bf9f877cc",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/15.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/15.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 16,
"slug": "cinematic-sci-fi-suspense-video-a-realistic-spaceship-crew-in-024495",
"title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "太空船门后的小学教室"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085498501384024495"
},
"source_prompt": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic uniforms follows a mysterious distress signal and boards a dark, abandoned spacecraft drifting in space. Flashlights cut through darkness as they move cautiously through narrow metallic corridors. Warning lights flicker, metal creaks, and their boots echo on the floor. One crew member says, \"Signal's coming from deeper inside.\" They enter a large chamber filled with hundreds of empty cryogenic pods, all open and abandoned. Another crew member says nervously, \"Where did everyone go?\" They approach a final closed door at the far end. The captain slowly opens it, and instead of another ship compartment, it reveals a completely normal present-day elementary school classroom in bright daylight. Children sit at desks, backpacks on the floor, colorful posters on the walls, and a teacher at the front turns to the crew and scolds them: \"You're late again. Take your seats.\" The astronauts stand frozen in confusion at the doorway. Realistic tone, smooth camera movement, strong suspense buildup, with the final reveal clear, sudden, and absurd. Include realistic audio: radio static, low ship hum, footsteps, pod machinery ambience, door hiss, then ordinary classroom sounds and the teacher's voice.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic science-fiction suspense, realistic spaceship crew in full futuristic uniforms boarding a dark abandoned spacecraft in response to a distress signal. A smooth tracking shot follows them through a narrow metallic corridor; flashlight beams cross flickering warning lights. One crew member with a restrained mid-pitched English voice (S1) says: [English] Signal's coming from deeper inside. Radio static, a low ship hum, creaking metal and hard boot echoes accompany their advance.\n[Shot 2] At 00:04.500, the camera cuts to a wide view of a large chamber with hundreds of open, empty cryogenic pods. A second crew member with an uneasy, slightly higher English voice (S2) looks from the abandoned pods toward a final closed door and says nervously: [English] Where did everyone go? The captain approaches that door as the camera pushes forward with the crew. Pod machinery resonates beneath their footsteps.\n[Shot 3] At 00:09.000, the camera cuts behind the captain as the door hisses open. Bright daylight reveals an ordinary present-day elementary school classroom, children at desks, backpacks on the floor and colorful posters on the walls. A teacher at the front with a clear, matter-of-fact English voice (S3) turns toward the astronauts and scolds them: [English] You're late again. Take your seats. The astronauts remain frozen in confusion at the doorway. Hold the absurd reveal until the end while ordinary classroom room tone replaces the ship ambience.\n\noverall_soundscape: Radio static, a low ship hum, creaking metal, boot echoes and empty pod machinery fill the spacecraft. A door hiss gives way abruptly to ordinary classroom chair movement and quiet paper rustling.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "515b6f9c02a9aff5ebdd48c593434b27848f0a398afc1c0651103dfdd475c0da",
"duration": 15,
"seed": 57,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.019365294836462,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/16.mp4",
"poster": "assets/v2-15s/lightx2v-4/16.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6812769,
"sha256": "74f38fb47041f298c2839e70c5dcbd72ecbb7324d1f313abff9009167998a207",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/16.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/16.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 17,
"slug": "skyship-flight-across-a-floating-kingdom-single-continu-344891",
"title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "穿越浮岛王国的飞艇"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@Strength04_X",
"url": "https://x.com/Strength04_X/status/2082712692159344891"
},
"source_prompt": "Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds. Complete the described path in one uninterrupted 15-second flight, establishing the skyship first and reserving the final seconds for the sunlight reveal beyond the storm.\n\noverall_soundscape: High-altitude wind and engine thrust deepen as the skyship enters thunder, lashing rain and cracking rock. Waterfalls rush past the hull before the storm sound falls behind the ship in open sunlight.\n\nnon_diegetic_music: Broad sustained strings and steady low orchestral percussion build through the storm, then open into a long brass-and-string chord in sunlight.",
"prompt_sha256": "15ed03a4ff9ad4d170616c8220d15f8c7ddadb16a059a522c567a1cede541838",
"duration": 15,
"seed": 58,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "飞空艇穿越浮空王国"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.037491537630558,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/17.mp4",
"poster": "assets/v2-15s/lightx2v-4/17.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 10047621,
"sha256": "087ed14d9efdd260f5d658f36b9323ce34f87ca864f119a92052a7f02cb7f5fa",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/17.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/17.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 18,
"slug": "a-hyper-realistic-handheld-phone-video-of-a-quiet-suburban-495564",
"title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "后院上空突然坠落的天空"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2086879654116495564"
},
"source_prompt": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\" Use a single 15-second home-video take. Preserve the first several seconds of ordinary watering, begin the overhead crack after that peaceful lead-in, and end with the sky descending and camera falling.\n\noverall_soundscape: Ordinary backyard birds, distant traffic and a watering stream continue for several seconds. A sudden deep overhead crack becomes a heavy descending rumble, followed by startled breathing and the phone hitting the ground.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "42f8695e30223aeca06566bf0476f7882751c30d3db90ad0136585d517c6ba26",
"duration": 15,
"seed": 59,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.118513136170805,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/18.mp4",
"poster": "assets/v2-15s/lightx2v-4/18.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8065659,
"sha256": "37441c49a03fb06169ddbfc0ec0fb8180e900f1faf59131d8805862a64b300d4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/18.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/18.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 19,
"slug": "also-has-an-impressive-understanding-of-differen-911311",
"title": {
"en": "also has an impressive understanding of different cultures",
"zh": "穿行爪哇市场的肉丸摊贩"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@junwatu",
"url": "https://x.com/junwatu/status/2084840152715911311"
},
"source_prompt": "Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism. Nearby women bargaining in Javanese (S1) and vendors calling prices in Javanese (S2) remain overlapping background voices. Follow the vendor in one continuous 15-second handheld take, keeping the steaming bowl and his careful steps legible; finish with serving the bowl and the brief power flicker.\n\noverall_soundscape: Dense covered-market bustle includes footsteps, bowl clatter, steaming food and sacks brushing clothing. Slow motorcycle engines pass among the stalls, followed by a brief fluorescent power buzz and the uninterrupted market activity.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "644ee708e360b5247f1f86b9510bc54be46b1ede81be1dfd8e283227dfab466f",
"duration": 15,
"seed": 60,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "also has an impressive understanding of different cultures",
"zh": "also has an impressive understanding of different cultures"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 14.480799765791744,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/19.mp4",
"poster": "assets/v2-15s/lightx2v-4/19.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9210509,
"sha256": "e92844ee9bcaed12bee7b32749331af0eef81bb85ecc3ac0d8df48a1ee2a8242",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/19.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/19.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 411.3973352322355,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 20,
"slug": "16-9-cinematic-wuxia-mystery-set-in-a-bamboo-571041",
"title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "夜雪竹林里的武侠悬疑"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@AIwithAliya",
"url": "https://x.com/AIwithAliya/status/2083132770650571041"
},
"source_prompt": "A 16:9 cinematic wuxia mystery set in a bamboo forest at night. Use a low-saturation palette of cold blue, ink green, charcoal, and gray. Thin mist fills the forest and fine snow drifts through the air. The mood is austere, lethal, and controlled, with the tension of a martial-arts sect investigating a case and exchanging secret intelligence.\n\nDeep in the forest, dense vertical bamboo fills the background while cold white mist-light glows in the distance. Soft, out-of-focus leaves partially obscure the foreground, creating the sense of watching from within the grove. A cool, soft front-side key lights the actors’ faces; backlight keeps the foreground dark and the distance luminous. Use shallow depth of field so leaves, snow, and bamboo dissolve into soft bokeh.\nPrioritize facial close-ups and measured shot/reverse-shot coverage. Keep the rhythm restrained but tense. Photoreal period-drama production value, cinematic lighting, and no modern elements. No subtitles, on-screen text, watermarks, modern clothing or architecture, animation styling, over-smoothed skin, bright daylight, or comic performance.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic wuxia mystery in a bamboo forest at night, in a restrained cold-blue, ink-green, charcoal and gray palette. Thin mist and fine drifting snow surround two period-clothed martial-arts investigators. Dense vertical bamboo fills the background; soft foreground leaves partially obscure a tight facial close-up of the first investigator. Cool, soft front-side light reveals natural skin texture while cold white mist-light glows in the distance. The investigator looks toward the other person with measured, wary attention. No modern objects, text, subtitles, watermarks or comic performance.\n[Shot 2] At 00:05.000, the camera cuts to the other investigator in matching close-up for a restrained reverse angle. Keep the same cold lighting, shallow depth of field and softly blurred falling snow. A small shift of the eyes toward the dark grove conveys the exchange of secret intelligence without adding invented dialogue. Clothing rustles quietly under the bamboo wind.\n[Shot 3] At 00:10.000, the camera cuts back to the first investigator seen through out-of-focus leaves. A slow, small push-in follows a controlled glance into the luminous mist. Hold the austere, tense expression until the end, with the distant forest remaining still except for snow and moving leaves. Photoreal period-drama lighting, no animation styling, bright daylight or over-smoothed skin.\n\noverall_soundscape: Thin wind passes through bamboo leaves with occasional dry stalk creaks. Soft period-clothing movement and faint footsteps remain close while falling snow and distant forest space stay quiet.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "11628027721bd267f57680bb10ff37aff3623b6ea6bbfb064178144173e3a828",
"duration": 15,
"seed": 61,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "竹林武侠谜案"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 15.904188649728894,
"observed_sparse_calls": 144,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-4/20.mp4",
"poster": "assets/v2-15s/lightx2v-4/20.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 2621884,
"sha256": "f7e0b03b55e9147092525f1e3ec0e4d69bdbb44cc0fe5a0e8ab160068ca904aa",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-4/20.mp4",
"dataset_path": "videos/v2-15s/lightx2v-4/20.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 1,
"slug": "via-mlx-serve-by-561487",
"title": {
"en": "via MLX Serve by",
"zh": "小恐龙打喷嚏变出蝴蝶"
},
"category": "animation",
"source": {
"kind": "x",
"name": "@albertgao",
"url": "https://x.com/albertgao/status/2085137397134561487"
},
"source_prompt": "\"A cute little girl walks through a sunny forest when a small, friendly dinosaur steps out, waves, and says hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style.\"\n\n- time: 15min ish\n- 768x768",
"source_duration": "5s",
"prompt": "integrated_multimodal_description: [Shot 1] A cute little girl walks through a sunny forest when a small, friendly dinosaur with a light, cheerful English voice (S1) steps out, waves, and says: [English] hi. Suddenly, the dinosaur sneezes—and a shower of colorful butterflies bursts from its nose. They both laugh. Cute, colorful animated style. Pace the meeting, sneeze and shared laughter over a continuous 15-second scene, allowing time to see the butterflies settle around both characters.\n\noverall_soundscape: Light forest wind, birds and small footsteps accompany the meeting. A comic sneeze releases a fluttering rush of butterfly wings, followed by the girl and dinosaur laughing together.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "954337465d95f9ab7b1fd512ba3879fa925f8e668f493e314c9d84a91e99fad3",
"duration": 15,
"seed": 42,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 5s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "via MLX Serve by",
"zh": "via MLX Serve by"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 29.708924832753837,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/01.mp4",
"poster": "assets/v2-15s/lightx2v-8/01.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4444968,
"sha256": "bb300ec0825a2ac8430d390e730d646e7caf697ed827934d120006de63060745",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/01.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/01.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 2,
"slug": "the-last-thing-you-see-in-your-first-and-last-290296",
"title": {
"en": "The last thing you see in your first and last",
"zh": "太空旅客窗外的黑洞"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@ivanfioravanti",
"url": "https://x.com/ivanfioravanti/status/2086553101029290296"
},
"source_prompt": "amateur handheld pov footage of a tourist in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: \"Wow look at that!\", then they show back the room",
"source_duration": "10s",
"prompt": "integrated_multimodal_description: [Shot 1] amateur handheld pov footage of a tourist with an excited, conversational English voice (S1) in their plush and comfortable room of a space cruise, carpeted floor and comfy bed, large window shows a view of Gargantua blackhole, you can see their reflection in the window, they turn off the light half way through so we can see outside better and say: [English] Wow look at that!, then they show back the room Use one uninterrupted 15-second handheld take. Switch off the room light around halfway through, hold the window view for the spoken reaction, then turn back to show the room.\n\noverall_soundscape: A steady enclosed-room ventilation hum and soft carpeted footsteps continue under slight handheld handling noise. A light-switch click marks the change to the darkened room.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "712f02d91f39b4d5a029cf4c3396609ac32d151511bf04949501f697139405e5",
"duration": 15,
"seed": 43,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 10s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "The last thing you see in your first and last",
"zh": "The last thing you see in your first and last"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.4092057608068,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/02.mp4",
"poster": "assets/v2-15s/lightx2v-8/02.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5588386,
"sha256": "93f9f21c99b7db75349f9ad37f089de929cee64a39fa46033c6b3c20760d9a20",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/02.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/02.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 3,
"slug": "ultra-cinematic-macro-shot-a-calm-mountain-lake-039920",
"title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "水滴汇成 MORNING"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2085622021812039920"
},
"source_prompt": "Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic macro shot: A calm mountain lake at dawn reflects the first golden rays of sunlight. Tiny ripples spread across the mirror-like surface as morning mist drifts gracefully above the water. Billions of sparkling droplets slowly rise into the air, suspended as if gravity has stopped. They spiral together with incredible fluid realism, naturally sculpting the word \"MORNING\" entirely from crystal-clear water. The liquid letters shimmer with golden reflections before gently collapsing into a spectacular explosion of droplets illuminated by the sunrise. Hollywood title sequence, ultra-realistic fluid simulation, volumetric lighting, IMAX quality. In one continuous 15-second camera move, establish the dawn lake first, raise and spiral the droplets in the middle, hold the completed readable word before its final liquid collapse.\n\noverall_soundscape: Quiet mountain wind, small lake ripples and faint distant dawn birds establish the lakeside. Delicate water movement grows into the rushing liquid-letter collapse and a crisp shower of droplets.\n\nnon_diegetic_music: Slow, sustained string notes and sparse high piano tones rise gently with the water lettering and resolve as the droplets fall.",
"prompt_sha256": "7abb51ba6a7327cb9266b13da9a8fa218928aeb0d97f444c82dc95ea21c245c2",
"duration": 15,
"seed": 44,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Ultra cinematic macro shot: A calm mountain lake at dawn",
"zh": "Ultra cinematic macro shot: A calm mountain lake at dawn"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.10105863492936,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/03.mp4",
"poster": "assets/v2-15s/lightx2v-8/03.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6564820,
"sha256": "7d0669c435502ed2c7a25c2771be534d7846656231602cab6540bfb3c6d650b7",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/03.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/03.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 4,
"slug": "strawberry-seasonal-match-cut-food-commercial-117943",
"title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓穿过四季落入奶油"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@HBCoop_",
"url": "https://x.com/HBCoop_/status/2082827829172117943"
},
"source_prompt": "Premium food commercial. Extreme macro shot of a single ripe strawberry falling toward a white ceramic bowl. Before it lands, the environment changes through four seamless seasonal match cuts: a sunlit spring greenhouse, a hot summer picnic, an autumn kitchen and a candlelit winter dining room. The strawberry stays in exactly the same screen position, size, orientation and downward motion through every transition. It lands in fresh cream during the winter scene, sending one graceful crown-shaped splash upward. Photorealistic food texture, high-speed product cinematography, clean commercial lighting. Each season has distinct environmental sound, but one continuous musical phrase connects the entire sequence. No hands, labels or additional fruit.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Photorealistic premium food commercial, extreme macro high-speed product cinematography. A single ripe strawberry falls toward a white ceramic bowl in a sunlit spring greenhouse. Its red seeds and moist surface are sharply resolved against soft green glasshouse bokeh. Keep the strawberry's screen position, size, orientation and downward motion consistent through every seasonal match cut. No hands, labels or additional fruit.\n[Shot 2] At 00:03.000, the shot transitions to a hot summer picnic in a seamless seasonal match cut around the same falling strawberry and bowl. Warm sunlight replaces the greenhouse light; distant insects accompany its uninterrupted descent.\n[Shot 3] At 00:06.000, the shot transitions to an autumn kitchen with amber window light, preserving the strawberry and bowl alignment and the same downward motion. Soft indoor room tone replaces the outdoor insects.\n[Shot 4] At 00:09.000, the shot transitions to a candlelit winter dining room. The strawberry completes its fall into fresh cream in the white bowl, sending one graceful crown-shaped splash upward. Hold the detailed splash and its settling ripples through the final seconds. One continuous musical phrase connects all four seasons.\n\noverall_soundscape: Spring birds and a gentle greenhouse breeze change to summer insects, quiet autumn kitchen ambience, then the still winter dining room. The final cream splash and falling droplets remain clearly audible.\n\nnon_diegetic_music: One uninterrupted light piano-and-plucked-string phrase runs at a moderate tempo across all four seasons, resolving with the strawberry splash.",
"prompt_sha256": "0c4614d6b1c88a86974b1393eab7a574817abc11c37bf72e3cb53e2d9bbf718f",
"duration": 15,
"seed": 45,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Strawberry seasonal match-cut food commercial",
"zh": "草莓四季匹配剪辑食品广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.147748040035367,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/04.mp4",
"poster": "assets/v2-15s/lightx2v-8/04.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1963781,
"sha256": "4e377794bc98d94006d5de362057d2b6e7edd605444af045c3fab0fea5f7ec20",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/04.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/04.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 5,
"slug": "stunning-action-packed-scene-following-a-little-robot-r-288619",
"title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "赛博城市里的机器人追逐"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@AIandDesign",
"url": "https://x.com/AIandDesign/status/2082522979339288619"
},
"source_prompt": "A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A stunning action packed scene following a little robot running away from soldiers, drones and other things that are trying to get him. Chase cam, dystopian futuristic cyberpunk environment, cinematic lighting. The camera follows the robot from behind, chase cam style. Every now and then the robot looks behind him and the entities chasing it are shown. Maintain one continuous 15-second rear chase shot with brief arcs that reveal the pursuers when the robot looks back; preserve the robot’s identity and forward motion.\n\noverall_soundscape: Rapid metallic footfalls and small robot servo movements run through the city ambience. Heavy pursuing footsteps, drone rotors and passing machinery track the chase without intelligible dialogue.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be75f61112b0d84692fb5ce91ced32ae50c3f1a3a1ad6b115fba2089944d4073",
"duration": 15,
"seed": 46,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Little Robot Cyberpunk Escape",
"zh": "小机器人赛博朋克逃亡"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.32196999480948,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/05.mp4",
"poster": "assets/v2-15s/lightx2v-8/05.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9521493,
"sha256": "2c52140f346b2b052a8e2dec2436cab24e57517d08abdde6671d2146192f8442",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/05.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/05.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 6,
"slug": "created-with-minimax-h3-max-120666",
"title": {
"en": "Created with MiniMax H3 Max",
"zh": "湖畔花束与风"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@TaliaAariz",
"url": "https://x.com/TaliaAariz/status/2095857062987120666"
},
"source_prompt": "A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A cinematic medium shot of a gentle East Asian young woman with dark hair styled in a bun adorned with a blue-and-white checkered bow, standing on a sunny pebbled lakefront. She wears a sleeveless light blue polka-dot tulle dress with long, translucent white ribbons flowing from her shoulder that flutter gracefully in a gentle breeze. Holding a lush bouquet of pastel pink, baby blue, and white flowers, she smiles warmly at the camera, softly touches the flower petals, and tilts her head back slightly to close her eyes and feel the breeze. In the background, glistening blue lake water sparkles under the sun, bordered by tall green reeds along the shore and serene blue mountains under a bright, clear sky, captured in soft natural daylight with a shallow depth of field. Keep one continuous 15-second medium shot, with a very slow small push-in as she touches the flowers, smiles and finally closes her eyes to feel the breeze.\n\noverall_soundscape: Gentle shore wind moves the ribbons and reeds while small waves wash over the pebbles. Quiet bouquet and fabric rustles accompany her hand movements.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7e05bce3aa444d99e753af9bcc4865bd2934e2c9102f5a1ce63f952302516d15",
"duration": 15,
"seed": 47,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3 Max",
"zh": "Created with MiniMax H3 Max"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.228797840885818,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/06.mp4",
"poster": "assets/v2-15s/lightx2v-8/06.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 4388971,
"sha256": "dc2fefcde62a6c513dcf62f2bdcc75c4247d2288885cbf453dfe81db5d1a99e4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/06.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/06.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 7,
"slug": "created-with-minimax-h3-on-989121",
"title": {
"en": "Created with MiniMax H3 on .",
"zh": "画家与湖畔城市天际线"
},
"category": "cinematic-travel",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087408671101989121"
},
"source_prompt": "A young Western artist sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western artist with a relaxed conversational English voice (S1) sets up an easel in a peaceful city park and paints the skyline overlooking a lake. As the artwork comes to life, she compares the finished painting with the real view, holding the canvas in front of the skyline and smiling proudly. Ultra-photorealistic visuals, cinematic documentary style, realistic brush strokes, authentic hand movements, natural English dialogue, handheld camera movement, detailed cityscape, premium lighting, seamless continuity, and immersive outdoor ambience. Her brief natural English remarks follow her painting actions. Keep the artist, easel and skyline consistent over the 15-second take. Begin with her positioning the easel, follow several visible finishing brush strokes, then give the final canvas-to-skyline comparison time to read.\n\noverall_soundscape: Soft city-park wind, distant traffic and lakeside birds surround the easel. Light wood movements, brush-on-canvas friction and cloth handling follow the artist’s actions.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "f39be188be70b922d6337aa1436e276c98592d5adf01475408b43bbd227e6cbc",
"duration": 15,
"seed": 48,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Created with MiniMax H3 on .",
"zh": "Created with MiniMax H3 on ."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.21258383290842,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/07.mp4",
"poster": "assets/v2-15s/lightx2v-8/07.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3948104,
"sha256": "9ed77d7a90bbe1912931c8d6781454641992bc26214a6831ab92cf9a61a6de53",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/07.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/07.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 8,
"slug": "created-with-minimax-h3-522089",
"title": {
"en": "Created with MiniMax H3.",
"zh": "街头摄影师、老人和小狗"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@aiwithaly",
"url": "https://x.com/aiwithaly/status/2087102541146522089"
},
"source_prompt": "A young Western female street photographer walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says “Look at that,” then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] A young Western female street photographer with a warm conversational English voice (S1) walks through a lively downtown street and notices an elderly man sitting outside a café with his small dog. She carefully composes the candid moment through her camera, captures the photo, then turns the camera toward the viewer to proudly show the shot she just took. She smiles, says: [English] Look at that, then continues walking through the city. Ultra-photorealistic visuals, natural handheld documentary movement, realistic camera interaction, authentic facial expressions, accurate hand movements, realistic dog behavior, natural daylight, cinematic depth of field, continuous character consistency, immersive city ambience, premium documentary realism. Over one continuous 15-second handheld shot, follow her noticing and photographing the café scene, then turn with her as she shows the camera display, speaks and continues walking.\n\noverall_soundscape: Layered downtown footsteps, distant traffic and soft café room spill continue beneath a camera shutter click. The dog shifts and lightly pants beside the seated man.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "be91933cc24647e7105ff883517384c8c83c2ccef42801f72555f509c467e319",
"duration": 15,
"seed": 49,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with MiniMax H3.",
"zh": "Created with MiniMax H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.277623313944787,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/08.mp4",
"poster": "assets/v2-15s/lightx2v-8/08.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9975230,
"sha256": "4a663ea277af887bb157f498af70c7c735f2ea4afa77fdbf1736f89cc6e513c4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/08.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/08.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 9,
"slug": "low-angle-fashion-tracking-film-062019",
"title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "小巷里的时装与歌声"
},
"category": "fashion",
"source": {
"kind": "x",
"name": "@Kiber_Alla",
"url": "https://x.com/Kiber_Alla/status/2083583963512062019"
},
"source_prompt": "Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. She sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra realistic cinematic video, shot on ARRI Alexa 35 with spherical lens, natural daylight, photorealistic, looks like real footage filmed on location, high-end fashion film aesthetic, extreme detail, realistic skin texture, real fabric physics, subtle film grain. Continuous low angle tracking shot, camera smoothly moving backward as a beautiful young woman with long dark wind-blown hair walks and performs down a narrow sun-drenched urban alley between old textured stone buildings under deep blue sky. She wears a flowing milky white asymmetrical off-shoulder top and high-waist wide-leg technical cotton culottes with a thin rope belt. The wide pants and fabric move naturally and dramatically with the wind and her body. The woman, with a light melodic singing voice (S1), sings with genuine joy, smiling and softly laughing, maintaining strong eye contact with the camera, while performing free, light dance elements — fluid arm gestures, body swaying, and elegant full turns around herself as she continues walking forward. Her movement is energetic yet effortless, full of lightness and positive energy. Cinematic natural lighting with strong sunlight and deep shadows, realistic motion blur, high dynamic range, masterpiece, best quality, no artificial look. She performs a wordless vocal melody. Keep the low-angle backward tracking shot continuous for the full 15 seconds. Leave enough space for a full turn and return to eye contact without cutting.\n\noverall_soundscape: Light footsteps on stone, wind through the alley and natural fabric movement accompany the performance. Her soft laughter and breaths are synchronized to the visible movement.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "7c2058cc5dc84a848b29169d61ecc607158e3ce4e68dd07d43622c10bfd34d91",
"duration": 15,
"seed": 50,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词或歌词；未编造逐字台词，歌唱使用无词旋律。"
],
"source_title": {
"en": "Low-Angle Fashion Tracking Film",
"zh": "低机位时尚跟拍短片"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.00326671404764,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/09.mp4",
"poster": "assets/v2-15s/lightx2v-8/09.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9065393,
"sha256": "93292f48af79012c7f029b1610c960b4b626e8b189af59879a9b1d2b8d283659",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/09.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/09.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 10,
"slug": "generated-a-french-bulldog-with-minimax-h3-182507",
"title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "法国斗牛犬闯障碍"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085448660612182507"
},
"source_prompt": "A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] A casual handheld smartphone video filmed by the dog's owner in an ordinary suburban backyard. A cute, stocky French bulldog excitedly runs through a small homemade agility course, awkwardly weaving between plastic poles, hopping over two low hurdles, scrambling through a short fabric tunnel, and finally bumping a little bell with its nose. Follow the complete obstacle sequence in one 15-second handheld take, with the bell contact visible in the final seconds.\n\noverall_soundscape: Backyard birds and distant neighborhood ambience surround quick paw taps and excited dog panting. Plastic poles rattle, the fabric tunnel scrapes, and a small bell rings once at the finish.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "b441f7d70cc0c7a54efccb096492e34839955225bc31781700cf560af438bc8d",
"duration": 15,
"seed": 51,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Generated a french bulldog with MiniMax H3",
"zh": "Generated a french bulldog with MiniMax H3"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.325014712288976,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/10.mp4",
"poster": "assets/v2-15s/lightx2v-8/10.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 29972293,
"sha256": "31f8a5f8e6ac2376f2cdfb2dc2719f5bae332cb55349b527d040c9b48ea6e052",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/10.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/10.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 11,
"slug": "created-with-hailuo-h3-182619",
"title": {
"en": "Created with Hailuo H3.",
"zh": "冰川崩塌组成 EXTINCTION"
},
"category": "title-sequence",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2090740724803182619"
},
"source_prompt": "Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Ultra cinematic title sequence: A colossal glacier fractures across an endless frozen ocean. Deep blue cracks spread for miles beneath translucent ice before the entire shelf collapses. Mountains of ice are launched into the sky, suspended in breathtaking slow motion. Every crystal catches the sunlight while frozen debris rotates around the camera. The crystalline storm assembles the word \"EXTINCTION\", carved from pure glacial ice with intricate frozen textures. Hairline fractures slowly spread through every letter until the title violently shatters into millions of glittering ice fragments. Photorealistic ice simulation, IMAX-scale environmental VFX. Over 15 seconds, establish the glacier, follow the expanding fractures and airborne ice, hold the completed readable ice title, then finish on its violent shatter.\n\noverall_soundscape: Deep frozen-ocean wind surrounds spreading ice creaks and resonant fractures. Heavy ice collapse becomes a crystalline debris storm, followed by the sharp final title shatter.\n\nnon_diegetic_music: Sustained low strings swell with the ice fracture, joined by deep percussion at the shelf collapse and final title shatter.",
"prompt_sha256": "4450f406537c05f4eaf4263bdbb5e1e4d60f9b8e253dbe76aac8d0be5e2b24bd",
"duration": 15,
"seed": 52,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Created with Hailuo H3.",
"zh": "Created with Hailuo H3."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.329117204993963,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/11.mp4",
"poster": "assets/v2-15s/lightx2v-8/11.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8513657,
"sha256": "7efb6225b1c2ed0f5271ab8bb8b8a5de528e9fa7aff2bf092dc8c23b7d38e538",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/11.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/11.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 12,
"slug": "shatter-695817",
"title": {
"en": "SHATTER.",
"zh": "玻璃碎片组成 SHATTER"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@CharaspowerAI",
"url": "https://x.com/CharaspowerAI/status/2096554023809695817"
},
"source_prompt": "Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Premium action-film title sequence beginning with a gigantic transparent glass monolith floating against complete blackness. A tiny projectile impacts the exact center and an intricate network of cracks instantly races across the entire surface. The camera pushes directly into one expanding fracture, traveling through the microscopic crystalline structure as tension builds everywhere. Suddenly the monolith violently detonates into millions of glass fragments. The camera performs a fast 360-degree orbital move through the suspended crystal storm as every shard reflects sharp beams of cinematic light. An invisible force begins pulling selected fragments together, constructing the enormous word \"SHATTER\" from overlapping razor-thin glass shards. The completed transparent typography refracts the entire environment like a giant prism before another pressure wave passes through it, producing thousands of secondary fractures. The letters explode outward directly toward the lens in spectacular ultra slow motion. Extremely detailed glass physics, realistic refraction and caustics, elegant brutality, premium Hollywood VFX. Keep the camera move continuous over 15 seconds: establish the floating monolith, enter the spreading fracture, orbit the suspended fragments, hold the completed title briefly, then finish on fragments expanding toward the lens.\n\noverall_soundscape: A projectile impact triggers fine traveling glass cracks and a swelling structural groan. Dense crystalline shattering and passing shards follow the orbit, with a second pressure-wave crack and outward burst at the end.\n\nnon_diegetic_music: A sparse low electronic pulse builds beneath bowed-metal tones, with two deep percussion hits at the main and final glass explosions.",
"prompt_sha256": "b198e1ddb0d1d5c79cab984ce06e7944ecd4efbca41f105960c7312c540dd572",
"duration": 15,
"seed": 53,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "SHATTER.",
"zh": "SHATTER."
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.027931007090956,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/12.mp4",
"poster": "assets/v2-15s/lightx2v-8/12.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8906760,
"sha256": "55e4aa88d12443b97ec5c6eb07d3ff9129352b281785f346b15da0e475bc0242",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/12.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/12.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 13,
"slug": "shot-opens-tight-on-a-beautiful-woman-face-wind-452380",
"title": {
"en": "Cliffside Descent to the Ocean",
"zh": "从悬崖人像到俯拍海浪"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@umesh_ai",
"url": "https://x.com/umesh_ai/status/2082700637444452380"
},
"source_prompt": "The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] The shot opens tight on a beautiful woman face, wind tearing at her coat as she stands on a jagged cliff. The camera eases over her shoulder and tilts down, following her line of sight. We descend past the crumbling lip into the void, the frame widening to reveal cliffs dropping away. The ocean roars into view, violent surf detonating against rock. The move continues into a high overhead, locking into an aerial top-down that frames the lone figure at the precipice as waves explode below. Perform the described camera path as one continuous 15-second move, ending on the high overhead view of the lone figure and exploding surf.\n\noverall_soundscape: Strong coastal wind drives the coat fabric. Deep ocean roar becomes clearer as the camera descends, with violent waves striking rock beneath the final overhead view.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "8a9fb882646cdb521b55ea777eea72c255e9b3863f257602d34d5bda4da3b879",
"duration": 15,
"seed": 54,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Cliffside Descent to the Ocean",
"zh": "悬崖下行至海边"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 21.941673386842012,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/13.mp4",
"poster": "assets/v2-15s/lightx2v-8/13.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9885251,
"sha256": "b97f41616ef6c9c604d21015046bbfcb6ca537dfc7891e6d12a55c5e0ca64311",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/13.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/13.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 14,
"slug": "ugc-style-skincare-video-featuring-a-realistic-young-wo-242986",
"title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "镜前护肤产品口播"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@ZaraIrahh",
"url": "https://x.com/ZaraIrahh/status/2083011066800242986"
},
"source_prompt": "UGC-style skincare video featuring a realistic young woman speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, vertical 9:16, social media ad quality, highly realistic natural skin texture, not overly retouched.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] UGC-style skincare video featuring a realistic young woman with a clear conversational English voice (S1) speaking directly to the camera in front of a bathroom mirror while holding a premium facial serum bottle. She smiles naturally, talks about her skincare routine, and demonstrates applying the serum to her face. Authentic, conversational presentation with genuine facial expressions and hand gestures, natural morning light, casual home bathroom setting, clean skincare aesthetic, healthy glowing skin, minimal makeup, smartphone-shot look, candid lifestyle beauty content, soft depth of field, luxury beauty brand feel, landscape 16:9, social media ad quality, highly realistic natural skin texture, not overly retouched. Her short natural English presentation stays about her skincare routine. Use a continuous 15-second landscape smartphone shot framed to keep both her face and the bottle clearly visible. She presents the serum, applies it to her face, and returns her attention to the camera.\n\noverall_soundscape: Quiet bathroom room tone, small glass-bottle handling sounds and soft hand-on-skin movements accompany the presentation. Keep breaths and clothing movement natural and close.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "61f30fc7218593117f92f6c7e46d55e03de10b31642a4122a69a9eba43d40b4c",
"duration": 15,
"seed": 55,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "Bathroom Mirror Skincare UGC Ad",
"zh": "浴室镜前护肤 UGC 广告"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 21.861816126853228,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/14.mp4",
"poster": "assets/v2-15s/lightx2v-8/14.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 3979216,
"sha256": "4d77b24dadb5f8d7184438f0f6f764ed31c8c0067197fe7835c7f91dd5d3e727",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/14.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/14.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 15,
"slug": "found-footage-film-with-335284",
"title": {
"en": "Horror Film Study 335284",
"zh": "渔夫接到诡异来电"
},
"category": "horror",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2084842607411335284"
},
"source_prompt": "\"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A distorted voice says, \"Finally. I’ve been trying to switch places with you.\" The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"Hyper-realistic found-footage horror video, filmed by a cheap action camera mounted inside a small fishing boat on a quiet lake. A lone fisherman violently reels in something heavy. He pulls up a muddy smartphone tangled in weeds. He answers, breathing hard. A low distorted English voice from the phone speaker (S1) says: [English] Finally. I’ve been trying to switch places with you. The phone flashes blinding white and the footage glitches. When the image returns, the phone shows the terrified fisherman trapped inside its screen, silently pounding on the glass. Behind him, an identical soaking-wet fisherman slowly climbs out of the lake into the boat, notices the camera, and reaches toward the lens. Abrupt cut to black. Raw handheld realism, imperfect autofocus, clipped audio, no music, no stylization.\" Keep the cheap boat-mounted camera viewpoint continuous for the 15-second scene, with only the specified white-flash glitch interrupting the image. Establish the fishing action, allow the phone line to finish, show the trapped image and then the wet double before the abrupt final black frame.\n\noverall_soundscape: Low lake water lapping, boat creaks and fishing-reel strain accompany the fisherman’s heavy breathing. Wet weeds scrape the hull, the phone glitches with clipped electronic noise, then water splashes as the soaked double enters the boat.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "778e3b3e9a6ca16d59312279338db624dd6281b72e5ade48402f63ed4b25f617",
"duration": 15,
"seed": 56,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "Horror Film Study 335284",
"zh": "恐怖短片案例 335284"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.16846158914268,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/15.mp4",
"poster": "assets/v2-15s/lightx2v-8/15.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 6226808,
"sha256": "384609c46099c2f747abbe61a659f1dd061a6229b6cb60bc151dc890c005ffcc",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/15.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/15.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 16,
"slug": "cinematic-sci-fi-suspense-video-a-realistic-spaceship-crew-in-024495",
"title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "太空船门后的小学教室"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2085498501384024495"
},
"source_prompt": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic uniforms follows a mysterious distress signal and boards a dark, abandoned spacecraft drifting in space. Flashlights cut through darkness as they move cautiously through narrow metallic corridors. Warning lights flicker, metal creaks, and their boots echo on the floor. One crew member says, \"Signal's coming from deeper inside.\" They enter a large chamber filled with hundreds of empty cryogenic pods, all open and abandoned. Another crew member says nervously, \"Where did everyone go?\" They approach a final closed door at the far end. The captain slowly opens it, and instead of another ship compartment, it reveals a completely normal present-day elementary school classroom in bright daylight. Children sit at desks, backpacks on the floor, colorful posters on the walls, and a teacher at the front turns to the crew and scolds them: \"You're late again. Take your seats.\" The astronauts stand frozen in confusion at the doorway. Realistic tone, smooth camera movement, strong suspense buildup, with the final reveal clear, sudden, and absurd. Include realistic audio: radio static, low ship hum, footsteps, pod machinery ambience, door hiss, then ordinary classroom sounds and the teacher's voice.",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic science-fiction suspense, realistic spaceship crew in full futuristic uniforms boarding a dark abandoned spacecraft in response to a distress signal. A smooth tracking shot follows them through a narrow metallic corridor; flashlight beams cross flickering warning lights. One crew member with a restrained mid-pitched English voice (S1) says: [English] Signal's coming from deeper inside. Radio static, a low ship hum, creaking metal and hard boot echoes accompany their advance.\n[Shot 2] At 00:04.500, the camera cuts to a wide view of a large chamber with hundreds of open, empty cryogenic pods. A second crew member with an uneasy, slightly higher English voice (S2) looks from the abandoned pods toward a final closed door and says nervously: [English] Where did everyone go? The captain approaches that door as the camera pushes forward with the crew. Pod machinery resonates beneath their footsteps.\n[Shot 3] At 00:09.000, the camera cuts behind the captain as the door hisses open. Bright daylight reveals an ordinary present-day elementary school classroom, children at desks, backpacks on the floor and colorful posters on the walls. A teacher at the front with a clear, matter-of-fact English voice (S3) turns toward the astronauts and scolds them: [English] You're late again. Take your seats. The astronauts remain frozen in confusion at the doorway. Hold the absurd reveal until the end while ordinary classroom room tone replaces the ship ambience.\n\noverall_soundscape: Radio static, a low ship hum, creaking metal, boot echoes and empty pod machinery fill the spacecraft. A door hiss gives way abruptly to ordinary classroom chair movement and quiet paper rustling.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "515b6f9c02a9aff5ebdd48c593434b27848f0a398afc1c0651103dfdd475c0da",
"duration": 15,
"seed": 57,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic",
"zh": "cinematic sci-fi suspense video. A realistic spaceship crew in full futuristic"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.159618015866727,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/16.mp4",
"poster": "assets/v2-15s/lightx2v-8/16.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5828206,
"sha256": "051fa4ea1d9615dfbc7276abb34bd1e7b28cc4647f2b1a0d762be56bd9e75796",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/16.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/16.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 17,
"slug": "skyship-flight-across-a-floating-kingdom-single-continu-344891",
"title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "穿越浮岛王国的飞艇"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@Strength04_X",
"url": "https://x.com/Strength04_X/status/2082712692159344891"
},
"source_prompt": "Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Skyship flight across a floating kingdom (single continuous shot) From above a sea of endless clouds, the camera dives toward a majestic skyship soaring between gigantic floating islands. Lock-on: the airship accelerating into a violent thunderstorm suspended in the sky. The camera slingshots beneath the hull, whips across the towering sails, then drops tight behind the glowing engines: lightning tears across the clouds, rain lashes the deck, shattered rock fragments tumble through the air. A collapsing floating island sends massive boulders crashing into the storm; the captain steers beneath a breaking stone arch before threading between waterfalls pouring into the sky, hanging bridges and drifting islands in one uninterrupted line. The camera mirrors every impossible maneuver before bursting through the storm into brilliant sunlight, revealing an endless kingdom of floating continents stretching beyond the horizon above a sea of clouds. Complete the described path in one uninterrupted 15-second flight, establishing the skyship first and reserving the final seconds for the sunlight reveal beyond the storm.\n\noverall_soundscape: High-altitude wind and engine thrust deepen as the skyship enters thunder, lashing rain and cracking rock. Waterfalls rush past the hull before the storm sound falls behind the ship in open sunlight.\n\nnon_diegetic_music: Broad sustained strings and steady low orchestral percussion build through the storm, then open into a long brass-and-string chord in sunlight.",
"prompt_sha256": "15ed03a4ff9ad4d170616c8220d15f8c7ddadb16a059a522c567a1cede541838",
"duration": 15,
"seed": 58,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Skyship Through a Floating Kingdom",
"zh": "飞空艇穿越浮空王国"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 21.763935210183263,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/17.mp4",
"poster": "assets/v2-15s/lightx2v-8/17.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 9926134,
"sha256": "794cbc5060de760017731b8e163c12e1d49304ce3ee560ad9898f2a2d3fc3bd0",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/17.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/17.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 18,
"slug": "a-hyper-realistic-handheld-phone-video-of-a-quiet-suburban-495564",
"title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "后院上空突然坠落的天空"
},
"category": "music-video",
"source": {
"kind": "x",
"name": "@cocktailpeanut",
"url": "https://x.com/cocktailpeanut/status/2086879654116495564"
},
"source_prompt": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\"",
"source_duration": "14s",
"prompt": "integrated_multimodal_description: [Shot 1] \"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright normal afternoon. A person casually waters plants near a fence while birds chirp, distant traffic hums, and everything feels completely ordinary and mundane. The camera drifts naturally like a real home video, capturing the lawn, patio furniture, and blue sky with soft white clouds. After several seconds of normal peaceful activity, a deep cracking sound suddenly comes from above. Without warning, the entire sky begins dropping straight downward like a gigantic solid ceiling, with the blue sky and clouds moving as one physical surface descending toward the yard. The person looks up in shock just as the sky rapidly fills the frame, swallowing the scene in a terrifying instant. The camera jerks and falls to the ground at the last moment. No cinematic buildup, no surreal visual style, it should feel like a totally normal real-life recording interrupted by one sudden impossible and visceral event.\" Use a single 15-second home-video take. Preserve the first several seconds of ordinary watering, begin the overhead crack after that peaceful lead-in, and end with the sky descending and camera falling.\n\noverall_soundscape: Ordinary backyard birds, distant traffic and a watering stream continue for several seconds. A sudden deep overhead crack becomes a heavy descending rumble, followed by startled breathing and the phone hitting the ground.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "42f8695e30223aeca06566bf0476f7882751c30d3db90ad0136585d517c6ba26",
"duration": 15,
"seed": 59,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源条目元数据为 14s，本次明确扩展为 15 秒。"
],
"source_title": {
"en": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright",
"zh": "\"A hyper-realistic handheld phone video of a quiet suburban backyard on a bright"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.57516465522349,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/18.mp4",
"poster": "assets/v2-15s/lightx2v-8/18.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8492778,
"sha256": "832f6c0b87fb6457b6f4e651b42f388838230ef03ed0dd6993bafd9301ce87e4",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/18.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/18.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 19,
"slug": "also-has-an-impressive-understanding-of-differen-911311",
"title": {
"en": "also has an impressive understanding of different cultures",
"zh": "穿行爪哇市场的肉丸摊贩"
},
"category": "cinematic-story",
"source": {
"kind": "x",
"name": "@junwatu",
"url": "https://x.com/junwatu/status/2084840152715911311"
},
"source_prompt": "Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Traditional Javanese market in Yogyakarta, early morning, present day, ultra-realistic handheld one-take. A bakso vendor hurries through impossibly narrow, crowded market aisles carrying a large steaming bowl of freshly prepared bakso. Shoppers and vendors step aside to let him pass. He weaves between vegetable stalls overflowing with chilies, shallots, and leafy greens, ducks beneath low-hanging tarps, squeezes past batik merchants and baskets of traditional snacks, and carefully avoids children running through the market. The market feels dense, humid, noisy, and full of life. He passes porters carrying heavy sacks on their shoulders, women bargaining in Javanese, and motorcycles slowly pushing through the narrow lane. He finally reaches a tiny food stall at the back of the market and serves the steaming bowl just as the power briefly flickers, leaving the market dim for a moment before the bustle continues. Oppressive realism, natural morning light mixed with fluorescent bulbs, worn concrete floors, damp walls, wooden stalls, hanging plastic signs, Javanese conversations, vendors calling out prices, clattering bowls, footsteps echoing through the covered market, immersive cinematic pacing. No cyberpunk, no stylization, documentary-like realism. Nearby women bargaining in Javanese (S1) and vendors calling prices in Javanese (S2) remain overlapping background voices. Follow the vendor in one continuous 15-second handheld take, keeping the steaming bowl and his careful steps legible; finish with serving the bowl and the brief power flicker.\n\noverall_soundscape: Dense covered-market bustle includes footsteps, bowl clatter, steaming food and sacks brushing clothing. Slow motorcycle engines pass among the stalls, followed by a brief fluorescent power buzz and the uninterrupted market activity.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "644ee708e360b5247f1f86b9510bc54be46b1ede81be1dfd8e283227dfab466f",
"duration": 15,
"seed": 60,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。",
"源文未提供固定台词；仅保留原文要求的语言与说话动作。"
],
"source_title": {
"en": "also has an impressive understanding of different cultures",
"zh": "also has an impressive understanding of different cultures"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.259512493852526,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/19.mp4",
"poster": "assets/v2-15s/lightx2v-8/19.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 8432796,
"sha256": "d5fa2b9596441bb344e6a14ca2021d85d14ac975a638bc857d920f7ac7677b01",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/19.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/19.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 433.3525244933553,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"id": 20,
"slug": "16-9-cinematic-wuxia-mystery-set-in-a-bamboo-571041",
"title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "夜雪竹林里的武侠悬疑"
},
"category": "product-commercial",
"source": {
"kind": "x",
"name": "@AIwithAliya",
"url": "https://x.com/AIwithAliya/status/2083132770650571041"
},
"source_prompt": "A 16:9 cinematic wuxia mystery set in a bamboo forest at night. Use a low-saturation palette of cold blue, ink green, charcoal, and gray. Thin mist fills the forest and fine snow drifts through the air. The mood is austere, lethal, and controlled, with the tension of a martial-arts sect investigating a case and exchanging secret intelligence.\n\nDeep in the forest, dense vertical bamboo fills the background while cold white mist-light glows in the distance. Soft, out-of-focus leaves partially obscure the foreground, creating the sense of watching from within the grove. A cool, soft front-side key lights the actors’ faces; backlight keeps the foreground dark and the distance luminous. Use shallow depth of field so leaves, snow, and bamboo dissolve into soft bokeh.\nPrioritize facial close-ups and measured shot/reverse-shot coverage. Keep the rhythm restrained but tense. Photoreal period-drama production value, cinematic lighting, and no modern elements. No subtitles, on-screen text, watermarks, modern clothing or architecture, animation styling, over-smoothed skin, bright daylight, or comic performance.",
"source_duration": "15s",
"prompt": "integrated_multimodal_description: [Shot 1] Live-action cinematic wuxia mystery in a bamboo forest at night, in a restrained cold-blue, ink-green, charcoal and gray palette. Thin mist and fine drifting snow surround two period-clothed martial-arts investigators. Dense vertical bamboo fills the background; soft foreground leaves partially obscure a tight facial close-up of the first investigator. Cool, soft front-side light reveals natural skin texture while cold white mist-light glows in the distance. The investigator looks toward the other person with measured, wary attention. No modern objects, text, subtitles, watermarks or comic performance.\n[Shot 2] At 00:05.000, the camera cuts to the other investigator in matching close-up for a restrained reverse angle. Keep the same cold lighting, shallow depth of field and softly blurred falling snow. A small shift of the eyes toward the dark grove conveys the exchange of secret intelligence without adding invented dialogue. Clothing rustles quietly under the bamboo wind.\n[Shot 3] At 00:10.000, the camera cuts back to the first investigator seen through out-of-focus leaves. A slow, small push-in follows a controlled glance into the luminous mist. Hold the austere, tense expression until the end, with the distant forest remaining still except for snow and moving leaves. Photoreal period-drama lighting, no animation styling, bright daylight or over-smoothed skin.\n\noverall_soundscape: Thin wind passes through bamboo leaves with occasional dry stalk creaks. Soft period-clothing movement and faint footsteps remain close while falling snow and distant forest space stay quiet.\n\nnon_diegetic_music: N/A",
"prompt_sha256": "11628027721bd267f57680bb10ff37aff3623b6ea6bbfb064178144173e3a828",
"duration": 15,
"seed": 61,
"adaptations": [
"保留社区原文及作者链接；本次按 H3 T2VA 三字段、镜头和语音规则整理。",
"按用户要求生成 15 秒（362 帧 / 24 FPS），画幅统一为 1344×768；新增环境音和镜头节奏均属于本次整理。"
],
"source_title": {
"en": "Bamboo Forest Wuxia Mystery",
"zh": "竹林武侠谜案"
},
"frames": 362,
"prompt_version": "v2-h3-15s",
"prompt_origin": "H3-format adaptation of attributed community source; not claimed to be the original author’s structured prompt",
"pipeline_s": 22.103271580766886,
"observed_sparse_calls": 336,
"video": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media/resolve/2d72924012f74288d6ce319ebfe915fa21162237/videos/v2-15s/lightx2v-8/20.mp4",
"poster": "assets/v2-15s/lightx2v-8/20.jpg",
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 2766861,
"sha256": "e8c29c07fde216be0a76f0028a57a49ee24caea50661a5fdbf27bb84d357cad1",
"verification": "ffprobe frame count and complete audio/video decode"
},
"local_video": "assets/v2-15s/lightx2v-8/20.mp4",
"dataset_path": "videos/v2-15s/lightx2v-8/20.mp4",
"media_dataset_revision": "2d72924012f74288d6ce319ebfe915fa21162237"
}
],
"benchmarks": [
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 404.07578102499247,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 37.770893225912005,
"repeats_s": [
37.68319828994572,
37.88577074185014,
38.53102756291628
],
"median_s": 37.88577074185014,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3532418,
"sha256": "fb524c8a5ffc3a4d63485ff7b9e58b5c9ce5fd52d39f449be410df377d7fec9a",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 404.07578102499247,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 71.94201591424644,
"repeats_s": [
71.8202538122423,
71.41925434302539,
71.52417825814337
],
"median_s": 71.52417825814337,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5503515,
"sha256": "a6a09f60e54a1029746dbb329c0a2598e69891ce1e36f56de59b099f9c1be00b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 404.07578102499247,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 29.702730207238346,
"repeats_s": [
14.27278988622129,
14.139453262090683,
14.144060876220465
],
"median_s": 14.144060876220465,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1439546,
"sha256": "0fd2c6240936b378cd40976fb5c34d09e2bdb1653324b45a87d57e5315127bc9",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 391.20716426102445,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 51.09772967081517,
"repeats_s": [
7.040904249064624,
7.030464249663055,
7.051090624183416
],
"median_s": 7.040904249064624,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 3,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20161,
"effective_density": 0.22213,
"route_count_min": 115,
"route_count_mean": 254.78,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3665198,
"sha256": "cc8777567182eac7e0962fce854141f29518479974f0b71ecd541e55989f8985",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 391.20716426102445,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 245.4264060538262,
"repeats_s": [
12.621026108041406,
12.619681135751307,
12.623293442185968
],
"median_s": 12.621026108041406,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1728,
"dense_calls": 672,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 3,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20161,
"effective_density": 0.22174,
"route_count_min": 160,
"route_count_mean": 378.07,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 600,
"dense_layer": 72
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5528454,
"sha256": "61dad109e536f05e0d29c26266e12391d76ca29a3d5c02b2749caf0bdf74de32",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 391.20716426102445,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 27.3790018921718,
"repeats_s": [
3.0292545007541776,
2.9962865291163325,
2.9951096647419035
],
"median_s": 2.9962865291163325,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20582,
"effective_density": 0.22908,
"route_count_min": 60,
"route_count_mean": 135.16,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1413144,
"sha256": "1417f5af3c2c2b283d6361b2b58db2d3f8ccb7ccb6bf36f6856d832d898867d3",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 398.27860447904095,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 10.703646476846188,
"repeats_s": [
3.7656888221390545,
3.7543425438925624,
3.752968884073198
],
"median_s": 3.7543425438925624,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 3,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18575,
"effective_density": 0.20664,
"route_count_min": 143,
"route_count_mean": 237.01,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3510338,
"sha256": "94d23d2ff334d76b883502f2ee5190f86a470301b7e3de530d67fab0c0d59b8d",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 398.27860447904095,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 243.63984071277082,
"repeats_s": [
6.691152920015156,
6.652609454933554,
6.675593917723745
],
"median_s": 6.675593917723745,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1728,
"dense_calls": 672,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 3,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18448,
"effective_density": 0.20552,
"route_count_min": 220,
"route_count_mean": 350.42,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 600,
"dense_layer": 72
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5590962,
"sha256": "88c646c9d73ccae4f10b8dc0421df69b99d7c56825ec6249d5455485e85c2d02",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 398.27860447904095,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 26.054769210983068,
"repeats_s": [
1.7181520899757743,
1.6623261668719351,
1.6656489758752286
],
"median_s": 1.6656489758752286,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18663,
"effective_density": 0.20946,
"route_count_min": 66,
"route_count_mean": 123.58,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1514911,
"sha256": "9d72b5fc5ee6b831d9c7a47827c19dd5b59d3ca554c7b325698f079ea1877393",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 425.4020757121034,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 65.33505730889738,
"repeats_s": [
65.48566196393222,
65.6046371506527,
65.33273641113192
],
"median_s": 65.48566196393222,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1572026,
"sha256": "be9021f12dc83658342a1574ade7038e90c277c680c798519f7e319843b7735a",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 425.4020757121034,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 129.85651914030313,
"repeats_s": [
129.51566505106166,
129.90104290517047,
129.23715650010854
],
"median_s": 129.51566505106166,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1919343,
"sha256": "7d26055f0c76b99c1f09bc9365060170a8ad680c7164daa06a07f3727c5c67a5",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 425.4020757121034,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 42.42403999203816,
"repeats_s": [
22.414969386998564,
22.482791357208043,
22.4277695142664
],
"median_s": 22.4277695142664,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 833903,
"sha256": "efa2a6e0dc9de037cb13d7657b3ff8fc249ab5c691fd7f95d835e16772968e3b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 408.11689551733434,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 71.77961420686916,
"repeats_s": [
71.96028976980597,
72.14366453094408,
71.87395159387961
],
"median_s": 71.96028976980597,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1635071,
"sha256": "32b29a029110098314abfe342730a5ad3c2501ad6b8f7d8d1c05be15351fa6ab",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 408.11689551733434,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 137.42840397078544,
"repeats_s": [
137.85603942908347,
137.42607980174944,
136.93979512900114
],
"median_s": 137.42607980174944,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1889995,
"sha256": "c8687863f2ed2460df95d6eb44c5534a5a57ae4de8cf3fdfbe0c36afb2784b4b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 1,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 408.11689551733434,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 40.22856629593298,
"repeats_s": [
25.559987880755216,
25.68099537305534,
25.446089582052082
],
"median_s": 25.559987880755216,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 932090,
"sha256": "88d411f81708016df471e2f0d46aa8a46968b9dec59c431649031cd3e12fef27",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 479.468451640103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 18.84309404483065,
"repeats_s": [
11.544127263128757,
11.55223888810724,
11.559273229911923
],
"median_s": 11.55223888810724,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20331,
"effective_density": 0.22398,
"route_count_min": 97,
"route_count_mean": 256.9,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1807659,
"sha256": "dae00a32244bf582679e80092cc171054e9ee4df1f8a235f892836c63f715979",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 479.468451640103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 28.802728682756424,
"repeats_s": [
21.485745965968817,
21.545224134344608,
21.510514890309423
],
"median_s": 21.510514890309423,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20332,
"effective_density": 0.22357,
"route_count_min": 158,
"route_count_mean": 381.19,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1934424,
"sha256": "3ca76c8c6af7724a6bb79045521999ae61b82d38a7b6a40574838c334c1c3421",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 479.468451640103,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 33.40463601704687,
"repeats_s": [
4.606777192093432,
4.593384526204318,
4.615335638169199
],
"median_s": 4.606777192093432,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20832,
"effective_density": 0.23173,
"route_count_min": 51,
"route_count_mean": 136.72,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 901555,
"sha256": "2ef7388b3b04cf6d9e61c0ab38192ac7a43c69651e9a8a4b5c8f7a5284525bb6",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 409.0616293284111,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 113.69129496766254,
"repeats_s": [
13.347808117046952,
13.34682912286371,
13.33125697216019
],
"median_s": 13.34682912286371,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20341,
"effective_density": 0.22407,
"route_count_min": 101,
"route_count_mean": 257.01,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1843870,
"sha256": "4aded58f1921b5870ea204051b6c73ef368c94b603f258ed0e6b2eee98628d0d",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 409.0616293284111,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 235.1847071829252,
"repeats_s": [
24.262370044831187,
24.287067759316415,
24.273042456246912
],
"median_s": 24.273042456246912,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20321,
"effective_density": 0.22349,
"route_count_min": 158,
"route_count_mean": 381.05,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1936290,
"sha256": "4d93e2878b88f9c775ab46f1df3233003f5fa5331742eac56a1c5c7df6926ea3",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 409.0616293284111,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 30.525569978170097,
"repeats_s": [
5.545859571546316,
5.51854232000187,
5.509740656707436
],
"median_s": 5.51854232000187,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.2084,
"effective_density": 0.23181,
"route_count_min": 52,
"route_count_mean": 136.77,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 926954,
"sha256": "6522c05d8ea33aebbb4be2cdea7ee5e122061509a6de1d5ab3e72fe12249e360",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 439.58678179606795,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 14.64438444795087,
"repeats_s": [
6.551806676667184,
6.531784829683602,
6.554025718010962
],
"median_s": 6.551806676667184,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18735,
"effective_density": 0.20817,
"route_count_min": 144,
"route_count_mean": 238.77,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1825597,
"sha256": "04260482aa6b6d7ec1550321d6dc7e0f04d7fa8cce53714711313f84ac47fc84",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 439.58678179606795,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 19.447643855120987,
"repeats_s": [
12.063323335722089,
12.064927336759865,
12.09263978432864
],
"median_s": 12.064927336759865,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18567,
"effective_density": 0.20671,
"route_count_min": 220,
"route_count_mean": 352.45,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1936513,
"sha256": "9fae8f6915cf26a28dbafda31f976c0cd4456c8b14f0a63ccc8b4010d00ac44b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3-int64",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 439.58678179606795,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 35.399448071140796,
"repeats_s": [
2.8105423622764647,
2.755053977947682,
2.758813951164484
],
"median_s": 2.758813951164484,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18678,
"effective_density": 0.20962,
"route_count_min": 63,
"route_count_mean": 123.67,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 915450,
"sha256": "13610b4257d37ebe6d83e9726d0194712b588c044c49bce154a17e68b6a7ddc8",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 405.47777480073273,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 47.894321171566844,
"repeats_s": [
7.4758722125552595,
7.466093507129699,
7.483112619724125
],
"median_s": 7.4758722125552595,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18758,
"effective_density": 0.2084,
"route_count_min": 141,
"route_count_mean": 239.03,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1786747,
"sha256": "b02db4baaf08434ef61616481171b10bb916eeab1db2291b763309091e2be215",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 405.47777480073273,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 266.72677598381415,
"repeats_s": [
13.31669664895162,
13.355944743845612,
13.337900619022548
],
"median_s": 13.337900619022548,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18555,
"effective_density": 0.20659,
"route_count_min": 218,
"route_count_mean": 352.24,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1909484,
"sha256": "9df91e67eb180ae2d0ba30865c89b5b44fc11df78cf0c83ffe207570f1e5c4e6",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "hsg",
"gpus": 8,
"allocated_gpus_per_node": "4",
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 405.47777480073273,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 143.39529226301238,
"repeats_s": [
3.241004516836256,
3.2079915329813957,
3.2244293242692947
],
"median_s": 3.2244293242692947,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18679,
"effective_density": 0.20964,
"route_count_min": 61,
"route_count_mean": 123.69,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 896271,
"sha256": "d152b689622c1ad2239157c2dbecafb11819034e5ce227a9715e229a2ce44233",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 382.99306421587244,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 27.033580739051104,
"repeats_s": [
2.964362938888371,
2.9692238410934806,
2.9639265229925513
],
"median_s": 2.964362938888371,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20726,
"effective_density": 0.23063,
"route_count_min": 54,
"route_count_mean": 136.07,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 655912,
"sha256": "419a740c383f13a3d0b9ef617ab32f710a59492f4b787779833c9634b260d612",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "hsg",
"gpus": 4,
"allocated_gpus_per_node": "4",
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 391.5484301429242,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 30.07582237990573,
"repeats_s": [
5.652423293795437,
5.53552816901356,
5.481880785897374
],
"median_s": 5.53552816901356,
"observed_sparse_calls": 336,
"sparse_stats": {
"sparse_calls": 1344,
"dense_calls": 256,
"sparse_calls_measured_request": 336,
"sparse_fraction": 0.84,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20782,
"effective_density": 0.23131,
"route_count_min": 49,
"route_count_mean": 136.47,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 56
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 772102,
"sha256": "39eceb078f98bfc967f27262f224b2b60e83b0cdb2539f48463debf049e11dda",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 286.39243665800313,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 37.879758728988236,
"repeats_s": [
37.825742068002,
37.85681245397427,
37.924753985978896
],
"median_s": 37.85681245397427,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3532418,
"sha256": "fb524c8a5ffc3a4d63485ff7b9e58b5c9ce5fd52d39f449be410df377d7fec9a",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 286.39243665800313,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 71.84661786601646,
"repeats_s": [
71.74971473100595,
71.8363306170213,
71.73549409999396
],
"median_s": 71.74971473100595,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5503515,
"sha256": "a6a09f60e54a1029746dbb329c0a2598e69891ce1e36f56de59b099f9c1be00b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 286.39243665800313,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 37.64529103299719,
"repeats_s": [
13.841052805015352,
13.813988408015575,
13.851194803020917
],
"median_s": 13.841052805015352,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1439546,
"sha256": "0fd2c6240936b378cd40976fb5c34d09e2bdb1653324b45a87d57e5315127bc9",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 294.7777604120056,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 14.31319376299507,
"repeats_s": [
7.229829151008744,
7.21938319200126,
7.224561576003907
],
"median_s": 7.224561576003907,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 3,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20161,
"effective_density": 0.22213,
"route_count_min": 115,
"route_count_mean": 254.78,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3665198,
"sha256": "cc8777567182eac7e0962fce854141f29518479974f0b71ecd541e55989f8985",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 294.7777604120056,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 55.12173893299769,
"repeats_s": [
12.985925676999614,
12.991896356994403,
12.980969693002407
],
"median_s": 12.985925676999614,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1728,
"dense_calls": 672,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 3,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20161,
"effective_density": 0.22174,
"route_count_min": 160,
"route_count_mean": 378.07,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 600,
"dense_layer": 72
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5528454,
"sha256": "61dad109e536f05e0d29c26266e12391d76ca29a3d5c02b2749caf0bdf74de32",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 294.7777604120056,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 29.798163390005357,
"repeats_s": [
3.0818069269880652,
3.0768804979888955,
3.075204674998531
],
"median_s": 3.0768804979888955,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20582,
"effective_density": 0.22908,
"route_count_min": 60,
"route_count_mean": 135.16,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1413144,
"sha256": "1417f5af3c2c2b283d6361b2b58db2d3f8ccb7ccb6bf36f6856d832d898867d3",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 285.7827172730031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 10.177076838008361,
"repeats_s": [
3.789100617010263,
3.7869828450056957,
3.783537929004524
],
"median_s": 3.7869828450056957,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 3,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18575,
"effective_density": 0.20664,
"route_count_min": 143,
"route_count_mean": 237.01,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 3510338,
"sha256": "94d23d2ff334d76b883502f2ee5190f86a470301b7e3de530d67fab0c0d59b8d",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 285.7827172730031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 12.688988487992901,
"repeats_s": [
6.722846845994354,
6.70660141800181,
6.693554066994693
],
"median_s": 6.70660141800181,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 1728,
"dense_calls": 672,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 3,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18448,
"effective_density": 0.20552,
"route_count_min": 220,
"route_count_mean": 350.42,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 600,
"dense_layer": 72
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 5590962,
"sha256": "88c646c9d73ccae4f10b8dc0421df69b99d7c56825ec6249d5455485e85c2d02",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "fasth3-4",
"adapter_sha256": "4ce198c83132251b7fd0de2503823aa49c53983f068318f66cb19eaefb7fcc12",
"nfe": 4,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 64,
"adapter_filename": "adapter_model.safetensors",
"cold_load_s": 285.7827172730031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 32.965851381988614,
"repeats_s": [
1.632835114010959,
1.6313898380030878,
1.6286788399884244
],
"median_s": 1.6313898380030878,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18663,
"effective_density": 0.20946,
"route_count_min": 66,
"route_count_mean": 123.58,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 1514911,
"sha256": "9d72b5fc5ee6b831d9c7a47827c19dd5b59d3ca554c7b325698f079ea1877393",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 294.1646487209946,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 63.84408454998629,
"repeats_s": [
63.75714229498408,
63.8483764519915,
63.88921740002115
],
"median_s": 63.8483764519915,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1574774,
"sha256": "da06db16d82e047ffb3fe684bb0ba0afee4be6b4691beac518e121e4f8fa413e",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 294.1646487209946,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 126.81416095001623,
"repeats_s": [
127.00113837901154,
127.09193216299172,
127.02317907201359
],
"median_s": 127.02317907201359,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1918724,
"sha256": "5a30dd3a3c5fb4c129429a665be1312db557762a113c971706fb92aaa7106212",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 294.1646487209946,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 30.864501964009833,
"repeats_s": [
21.856848203984555,
21.847355916979723,
21.877207930985605
],
"median_s": 21.856848203984555,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 832822,
"sha256": "c4f648d03ccbdab9a6b31b4086c3cb02ae5d6c5fc7ebb9b123549aba8728647b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 290.6031925349962,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 70.50726453700918,
"repeats_s": [
70.32355213700794,
70.31223342899466,
70.11613318399759
],
"median_s": 70.31223342899466,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1635071,
"sha256": "32b29a029110098314abfe342730a5ad3c2501ad6b8f7d8d1c05be15351fa6ab",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 290.6031925349962,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 135.63105801402708,
"repeats_s": [
135.65047668898478,
135.54529769800138,
135.5381122569961
],
"median_s": 135.54529769800138,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1889995,
"sha256": "c8687863f2ed2460df95d6eb44c5534a5a57ae4de8cf3fdfbe0c36afb2784b4b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 1,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "dense",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 290.6031925349962,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 71.83553511201171,
"repeats_s": [
24.93977749699843,
24.9366850979859,
24.955815777997486
],
"median_s": 24.93977749699843,
"observed_sparse_calls": 0,
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 932090,
"sha256": "88d411f81708016df471e2f0d46aa8a46968b9dec59c431649031cd3e12fef27",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 288.39288462500554,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 18.310874960006913,
"repeats_s": [
11.83686811098596,
11.855843441007892,
11.825516913988395
],
"median_s": 11.83686811098596,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20331,
"effective_density": 0.22398,
"route_count_min": 97,
"route_count_mean": 256.9,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1807659,
"sha256": "dae00a32244bf582679e80092cc171054e9ee4df1f8a235f892836c63f715979",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 288.39288462500554,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 28.251016705005895,
"repeats_s": [
22.0627290699922,
22.035492566996254,
22.032053962000646
],
"median_s": 22.035492566996254,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20332,
"effective_density": 0.22357,
"route_count_min": 158,
"route_count_mean": 381.19,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1934424,
"sha256": "3ca76c8c6af7724a6bb79045521999ae61b82d38a7b6a40574838c334c1c3421",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 288.39288462500554,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 21.500160736002726,
"repeats_s": [
4.695578521001153,
4.63270866998937,
4.648682238010224
],
"median_s": 4.648682238010224,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20832,
"effective_density": 0.23173,
"route_count_min": 51,
"route_count_mean": 136.72,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 901555,
"sha256": "2ef7388b3b04cf6d9e61c0ab38192ac7a43c69651e9a8a4b5c8f7a5284525bb6",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 287.7437038460048,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 47.6237547690107,
"repeats_s": [
13.509233238000888,
13.580609458003892,
13.604155460983748
],
"median_s": 13.580609458003892,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.20341,
"effective_density": 0.22407,
"route_count_min": 101,
"route_count_mean": 257.01,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1843870,
"sha256": "4aded58f1921b5870ea204051b6c73ef368c94b603f258ed0e6b2eee98628d0d",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 287.7437038460048,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 59.1786147270177,
"repeats_s": [
24.773393165989546,
24.86349293999956,
24.796716023003682
],
"median_s": 24.796716023003682,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.20321,
"effective_density": 0.22349,
"route_count_min": 158,
"route_count_mean": 381.05,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1936290,
"sha256": "4d93e2878b88f9c775ab46f1df3233003f5fa5331742eac56a1c5c7df6926ea3",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 287.7437038460048,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 64.3763722970034,
"repeats_s": [
5.503104650997557,
5.514375450991793,
5.521031700016465
],
"median_s": 5.514375450991793,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.2084,
"effective_density": 0.23181,
"route_count_min": 52,
"route_count_mean": 136.77,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 926954,
"sha256": "6522c05d8ea33aebbb4be2cdea7ee5e122061509a6de1d5ab3e72fe12249e360",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 292.3347832650179,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 13.000249734002864,
"repeats_s": [
6.537904661992798,
6.551716292015044,
6.550389580981573
],
"median_s": 6.550389580981573,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18735,
"effective_density": 0.20817,
"route_count_min": 144,
"route_count_mean": 238.77,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1825597,
"sha256": "04260482aa6b6d7ec1550321d6dc7e0f04d7fa8cce53714711313f84ac47fc84",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 292.3347832650179,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 18.49491580898757,
"repeats_s": [
12.103580157010583,
12.09581659201649,
12.15989164399798
],
"median_s": 12.103580157010583,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18567,
"effective_density": 0.20671,
"route_count_min": 220,
"route_count_mean": 352.45,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1936513,
"sha256": "9fae8f6915cf26a28dbafda31f976c0cd4456c8b14f0a63ccc8b4010d00ac44b",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "mxfp8",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 292.3347832650179,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 30.892118582007242,
"repeats_s": [
2.7030635649862234,
2.7144765239791013,
2.7327462660032324
],
"median_s": 2.7144765239791013,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18678,
"effective_density": 0.20962,
"route_count_min": 63,
"route_count_mean": 123.67,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 915450,
"sha256": "13610b4257d37ebe6d83e9726d0194712b588c044c49bce154a17e68b6a7ddc8",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 286.3570104090031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 10,
"frames": 243,
"seed": 20260903,
"warmup_s": 15.531070279015694,
"repeats_s": [
7.364248292986304,
7.369295509008225,
7.375663626007736
],
"median_s": 7.369295509008225,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 2304,
"dense_calls": 896,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 8,
"last_step": 7,
"video_start": 823,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 823,
"sink_blocks": null,
"sequence_length": 73399,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1147,
"sink_blocks": 13,
"threshold_density": 0.18758,
"effective_density": 0.2084,
"route_count_min": 141,
"route_count_mean": 239.03,
"route_count_max": 1147,
"metadata_capacity": 1152
},
"gate": null,
"declined": {
"warmup_step": 800,
"dense_layer": 96
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 243,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 10.208344,
"bytes": 1786747,
"sha256": "b02db4baaf08434ef61616481171b10bb916eeab1db2291b763309091e2be215",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 286.3570104090031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 15,
"frames": 362,
"seed": 20260903,
"warmup_s": 50.19084989000112,
"repeats_s": [
13.196016611997038,
13.217540967016248,
13.247293635999085
],
"median_s": 13.217540967016248,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 3456,
"dense_calls": 1344,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 12,
"last_step": 7,
"video_start": 1219,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 1219,
"sink_blocks": null,
"sequence_length": 109075,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 1705,
"sink_blocks": 20,
"threshold_density": 0.18555,
"effective_density": 0.20659,
"route_count_min": 218,
"route_count_mean": 352.24,
"route_count_max": 1705,
"metadata_capacity": 1728
},
"gate": null,
"declined": {
"warmup_step": 1200,
"dense_layer": 144
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 362,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 15.166666,
"bytes": 1909484,
"sha256": "9df91e67eb180ae2d0ba30865c89b5b44fc11df78cf0c83ffe207570f1e5c4e6",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "hyperflow-8",
"adapter_sha256": "4d7dec1363ebcb9fd63117621b65f8bd19fecacf7ba41f38dd098be363d3972d",
"nfe": 8,
"cluster": "poly",
"gpus": 8,
"allocated_gpus_per_node": null,
"nodes": 2,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 12.0,
"audio_shift": 3.0,
"adapter_alpha": 256.0,
"adapter_filename": "minimax_h3_hyperflow_8step_v1.0.safetensors",
"cold_load_s": 286.3570104090031,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 62.63353594500222,
"repeats_s": [
3.1236446789989714,
3.1219848209875636,
3.12078181700781
],
"median_s": 3.1219848209875636,
"observed_sparse_calls": 288,
"sparse_stats": {
"sparse_calls": 1152,
"dense_calls": 448,
"sparse_calls_measured_request": 288,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 2,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.18679,
"effective_density": 0.20964,
"route_count_min": 61,
"route_count_mean": 123.69,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 400,
"dense_layer": 48
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 896271,
"sha256": "d152b689622c1ad2239157c2dbecafb11819034e5ce227a9715e229a2ce44233",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "lightx2v-4",
"adapter_sha256": "c3d4a2cf618efea71b9e21a4baaa12d412f1eb6c2b6f86efacaf0ebb6814b689",
"nfe": 4,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_4step_v1.2_768p_bf16.safetensors",
"cold_load_s": 296.27685292600654,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 28.435087569989264,
"repeats_s": [
2.92887511299341,
2.925522841978818,
2.9275320450251456
],
"median_s": 2.9275320450251456,
"observed_sparse_calls": 144,
"sparse_stats": {
"sparse_calls": 576,
"dense_calls": 224,
"sparse_calls_measured_request": 144,
"sparse_fraction": 0.72,
"requests": 4,
"last_step": 3,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 200,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 200,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20726,
"effective_density": 0.23063,
"route_count_min": 54,
"route_count_mean": 136.07,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 24
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 655912,
"sha256": "419a740c383f13a3d0b9ef617ab32f710a59492f4b787779833c9634b260d612",
"verification": "ffprobe frame count and complete audio/video decode"
}
},
{
"adapter": "lightx2v-8",
"adapter_sha256": "9b0efe3613b43a84e30febaa43af27432ea9d0711eac7bba904b2556b175f6d4",
"nfe": 8,
"cluster": "poly",
"gpus": 4,
"allocated_gpus_per_node": null,
"nodes": 1,
"gpu": "NVIDIA GB200",
"capability": [
10,
0
],
"runtime_variant": "Sol-H3",
"torch": "2.10.0+cu130",
"cuda": "13.0",
"cudnn": 91501,
"attention_backend": "sol_bsa",
"compute_quant": "none",
"video_shift": 6.0,
"audio_shift": 3.0,
"adapter_alpha": 8,
"adapter_filename": "minimax_h3_fl2v_turbo_8step_v1.0_768p_bf16.safetensors",
"cold_load_s": 280.33616568100115,
"base_model_revision": "83db0c0efe6ef9824e0e194be110346c0a9542ed",
"width": 1344,
"height": 768,
"fps": 24,
"timing": "warm pipeline: text encoding + DiT + video/audio decoding; excludes load, warmup and MP4 encoding",
"duration": 5,
"frames": 124,
"seed": 20260903,
"warmup_s": 33.67662795999786,
"repeats_s": [
5.112099753008806,
5.099165264997282,
5.1182629440008895
],
"median_s": 5.112099753008806,
"observed_sparse_calls": 336,
"sparse_stats": {
"sparse_calls": 1344,
"dense_calls": 256,
"sparse_calls_measured_request": 336,
"sparse_fraction": 0.84,
"requests": 4,
"last_step": 7,
"video_start": 427,
"sink_mode": "prefix",
"sink_start": 0,
"sink_tokens": 427,
"sink_blocks": null,
"sequence_length": 37723,
"tau": 1.0,
"thresh_type": "diag",
"backend": "sol_bsa",
"qkv_wire_dtype": "int8_qkv",
"qkv_wire_policy": "int8_qkv_all",
"qkv_int8_calls_measured_request": 400,
"qkv_bf16_calls_measured_request": 0,
"qkv_wire_ratio": 0.520833,
"output_wire_dtype": "fp8",
"output_int8_calls_measured_request": 0,
"output_fp8_calls_measured_request": 400,
"output_bf16_calls_measured_request": 0,
"output_wire_ratio": 0.5,
"attention_comm_ratio": 0.515625,
"packed_input": true,
"compile_bucket_size": 4096,
"dense_steps": 1,
"dense_layers": 2,
"route_density": {
"blocks": 590,
"sink_blocks": 7,
"threshold_density": 0.20782,
"effective_density": 0.23131,
"route_count_min": 49,
"route_count_mean": 136.47,
"route_count_max": 590,
"metadata_capacity": 640
},
"gate": null,
"declined": {
"warmup_step": 200,
"dense_layer": 56
}
},
"media_validation": {
"video_codec": "h264",
"audio_codec": "aac",
"frames": 124,
"width": 1344,
"height": 768,
"fps": 24,
"sample_rate": 32000,
"channels": 2,
"duration_s": 5.258344,
"bytes": 772102,
"sha256": "39eceb078f98bfc967f27262f224b2b60e83b0cdb2539f48463debf049e11dda",
"verification": "ffprobe frame count and complete audio/video decode"
}
}
],
"source_revision": "03a0d24259341413597b3c6ff0ce0c0164007e8b",
"updated_at": "2026-09-13T09:37:36.623351+00:00",
"gallery_profile": {
"cluster": "hsg",
"active_gpus": 4,
"compute_quant": "none"
},
"runtime_code_commit": "bcce779cbe6bd7aa46d92adb4182136fc90a5724",
"cluster_status": {
"hsg": "Completed: 80 gallery videos and 23 benchmark configurations verified",
"poly": "Completed: 23 benchmark configurations verified; all final jobs completed 0:0"
},
"comparison_note": "HSG FastH3: eight of nine configurations are 0.19–2.90% slower than the published B300 references; one-GPU 15s is 36.86% slower. Hardware and benchmark prompt differ, so this is not exact same-hardware reproduction.",
"poly_comparison_note": "Poly FastH3 除单卡 15 秒外，相对原 B300 耗时差异为 -1.28% 至 +5.43%。单卡 15 秒耗时 71.750 秒，比原参照慢 37.29%；两个 GB200 集群都复现了这一差距。",
"gallery_revision": {
"version": "v2-h3-15s",
"status": "complete",
"duration_s": 15,
"frames": 362,
"prompt_validation": "H3-format rewrite of attributed source; full original text retained"
},
"media_repository": {
"repo_id": "Lawrence-cj/sol-h3-hyperflow-20260913-media",
"latest_commit": "87eca1622a36826aaad99a68a8556cf89e50ecb2",
"url": "https://huggingface.co/datasets/Lawrence-cj/sol-h3-hyperflow-20260913-media"
}
}