What is this?
An end-to-end, chat-driven, never-ending AI video broadcast: viewers type !prompt in Twitch or YouTube chat, an LLM expands the idea into a styled scene, the FastH3 model renders it as a 768p clip with synced audio, and it goes out continuously over RTMP.
An end-to-end, chat-driven, never-ending AI video broadcast: viewers type !prompt in Twitch or YouTube chat, an LLM expands the idea into a styled scene, the FastH3 model renders it as a 768p clip with synced audio, and it goes out continuously over RTMP.Is this one app or two separate pieces?It is split into two halves: fast-h3/ runs the video generation model and needs the Reactor Runtime with 8x B200 GPUs, while streaming-client/ handles chat ingestion, prompt upsampling and paced RTMP output and only needs Python with reactor-sdk plus ffmpeg. They communicate through a shared contract defined in fasth3types.py.
It is split into two halves: fast-h3/ runs the video generation model and needs the Reactor Runtime with 8x B200 GPUs, while streaming-client/ handles chat ingestion, prompt upsampling and paced RTMP output and only needs Python with reactor-sdk plus ffmpeg. They communicate through a shared contract defined in fasth3types.py.How do I get it running?First serve the model, either locally with `reactor run` from fast-h3/ or by deploying it, then start the streaming-client which pulls chat commands, queues scene generations against the model, and pushes the paced output over RTMP.
First serve the model, either locally with `reactor run` from fast-h3/ or by deploying it, then start the streaming-client which pulls chat commands, queues scene generations against the model, and pushes the paced output over RTMP.What hardware do I need to run the model?The model expects the Reactor Runtime on 8x B200 GPUs to serve FastH3 Preview v1; the streaming client itself just needs Python with reactor-sdk and any box with ffmpeg installed.
The model expects the Reactor Runtime on 8x B200 GPUs to serve FastH3 Preview v1; the streaming client itself just needs Python with reactor-sdk and any box with ffmpeg installed.Can I run the model on a regular single-GPU box?The model is built around the Reactor Runtime with 8x B200 GPUs, and the detailed weights layout, GPU/CUDA prerequisites and performance profile are documented separately in fast-h3/README.md rather than in this top-level README.
The model is built around the Reactor Runtime with 8x B200 GPUs, and the detailed weights layout, GPU/CUDA prerequisites and performance profile are documented separately in fast-h3/README.md rather than in this top-level README.What is FastH3 Preview v1?It is MiniMax-H3 (35B parameters) distilled by the FastVideo project down to four transformer forwards with 90% sparse video attention, generating video and audio jointly directly from text.
It is MiniMax-H3 (35B parameters) distilled by the FastVideo project down to four transformer forwards with 90% sparse video attention, generating video and audio jointly directly from text.Where do the stream's prompts actually come from?Viewers type !prompt directly in Twitch or YouTube chat, and an LLM expands each idea into a styled sequence of scenes before it is queued for generation.
Viewers type !prompt directly in Twitch or YouTube chat, and an LLM expands each idea into a styled sequence of scenes before it is queued for generation.Is there a minimal client to test the model queue directly?Yes, fast-h3/client/README.md documents a minimal reactor-sdk smoke-test client that walks through the raw queue contract once, without the full streaming pipeline.
Yes, fast-h3/client/README.md documents a minimal reactor-sdk smoke-test client that walks through the raw queue contract once, without the full streaming pipeline.What are similar projects?FreeVideo, hyperflow, MiniMax-H3, ComfyUI-MiniMax-H3-LongMedia, CodeVideoModel, vllm-omni, sglang, ComfyUI-ALLinONE-MinimaxH3
FreeVideo, hyperflow, MiniMax-H3, ComfyUI-MiniMax-H3-LongMedia, CodeVideoModel, vllm-omni, sglang, ComfyUI-ALLinONE-MinimaxH3Which topics is it in?knowledge-distillation, rtmp, minimax-h3, livestream, minimax, ai-video, inference-engine, audio-generation, webrtc, text-to-video, pretrained-models, youtube
knowledge-distillation, rtmp, minimax-h3, livestream, minimax, ai-video, inference-engine, audio-generation, webrtc, text-to-video, pretrained-models, youtube