FastVideo publishes ComfyUI weights for FastH3 Omni
Original title: FastVideo/FastH3-Omni-Comfy — new model trending #30 on Hugging Face
BF16 and INT8 ConvRot checkpoints use eight steps and need MiniMax-H3 text, video and audio VAEs.
FastVideo released ComfyUI-ready BF16 and INT8 ConvRot diffusion models for the final FastH3 Omni release. They generate video and synchronized audio from a prompt plus ordered image, video and audio references. Users install one diffusion checkpoint, a Qwen3-VL text encoder and MiniMax-H3 video and audio VAEs. Accelerated NVFP4 text encoding requires Blackwell hardware, and the weights are under the MiniMax-H3 Community License.
- Recommended sampling is eight steps with guidance 1.0.
- Video shift is 12.0 and audio shift is 3.0.
- The INT8 file is fasth3_omniref_8step_pruned_int8_convrot.safetensors.
- The audio VAE is required for synchronized audio.
- Nuva Lab co-developed the model and curated the reference dataset.
- Products
- ComfyUIMiniMax-H3
- AI models
- FastH3 Omni
From the original article
2 min read at huggingface.coComfyUI-ready BF16 and INT8 ConvRot diffusion models for the final FastH3 Omni release. Generate video and synchronized audio from a prompt and ordered image, video and audio references. [Release collection](https://huggingface.
By FastVideo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.