FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree · Hugging Face

Source: https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree

https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree#fastvideo-fasth3-4-step-preview-v1-vsa-datafreeFastVideo-FastH3-4-step-Preview-v1-VSA-DataFree

The recommended FastH3 Preview v1 checkpoint fromFastVideo. It generates synchronized video and audio from text with four transformer forwards. This step-1300 model was trained with data-free DMD2 and VSA-H3 at 90% sparsity.

Blog·Matching LoRA·FastH3 collection

This checkpoint requires FastVideo’s VSA-H3 attention backend. Use the matching LoRA above if you prefer to download only the distilled adapter.

https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree#run-with-fastvideoRun with FastVideo

Installuv, then use the CUDA 13 / Blackwell path below. It selects FastVideo’s published CUDA kernel wheel instead of compiling the kernel locally. See theinstallation guidefor other platforms.

git clone https://github.com/hao-ai-lab/FastVideo.git
cd FastVideo
uv venv --python 3.12 --seed
source .venv/bin/activate
UV_TORCH_BACKEND=cu130 uv pip install \
  --no-sources-package fastvideo-kernel \
  -e ".[fasth3]"

python examples/inference/basic/basic_fasth3.py \
  --model-path FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree \
  --prompt "your prompt" \
  --no-warmup \
  --repeats 1

The tested defaults use four B200 GPUs and the trained four-forward schedule. On other multi-GPU CUDA systems, follow the installation guide and add\-\-no\-replicated\-dit \-\-vsa\-kernel triton \-\-no\-fa4. The GPU count must divide H3’s 56 attention heads.

https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree#scopeScope

This preview supports text-to-audio-video generation. FL2VA and Ref2VA were not distilled. Difficult motion, fine detail, and some audio may remain below the base MiniMax H3 model. This checkpoint inherits theMiniMax H3 Community License.

https://huggingface.co/FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree#acknowledgementsAcknowledgements

We thankNuva Labfor bringing production grounding to FastH3 through its experience with real-world creative video-agent workloads. Its production-aligned post-training insights help bridge open-source research to practical data-assisted distillation for commercial video workflows, with Omni Ref as the next focus.

We thank theNVIDIA FastGenteam for theDMD2framework and H3 reference experiment that helped us align the score clock, modality shifts, and backward simulation.

We also thankMiniMaxfor releasing H3-Base, and thevLLM project,NVIDIA, andMBZUAIfor their continued sponsorship and support ofFastVideo.