← sources

FastH3 Omni is out with day-0 ComfyUI support: open weights, up to 11.9x faster reference-to-video

source: reddit (reddit:comfyui)published 2026-10-20 20:16ingested 2026-10-21 02:40

retained text

We just open-sourced FastH3 OmniRef from the FastVideo team at Hao AI Lab, and it works in ComfyUI from day one. Also available in the last release [0.42.2](http://0.42.in/) workflow templates

\- Feed it a prompt plus your own references (images, short video clips, a voice) and get a finished video with synced sound  
\- 8 steps instead of 50, up to 11.9x faster than base MiniMax-H3  
\- Open weights + training code

Links:  
\- Models: [https://huggingface.co/collections/FastVideo/fasth3-omni](https://huggingface.co/collections/FastVideo/fasth3-omni)  
\- Blog: [https://haoailab.com/blogs/fastvideo-omniref-update/](https://haoailab.com/blogs/fastvideo-omniref-update/)  
\- Code: [https://github.com/hao-ai-lab/FastVideo](https://github.com/hao-ai-lab/FastVideo)

Big thanks to NVIDIA, Nuva Lab, Reactor, and the ComfyUI team for helping make this happen. Open source must win!

Full local blog drops tomorrow. Would love to see what you build with it, drop your workflows below!

document fetched 2026-10-21 00:41 · canonical https://reddit.com/r/comfyui/comments/1xb2xz6/fasth3_omni_is_out_with_day0_comfyui_support_open