2026-10-23

FastH3-Omni open weights: image/video/audio references to short video with synced audio in one pass

On October 20, 2026, Nuva Lab and FastVideo (Hao AI Lab at UC San Diego) released FastH3-Omni: an open-weight, few-step acceleration of MiniMax H3 with full omni reference input (image, video and audio, in the same class of workflow as Seedance 2.5), producing about 5–15 seconds of 24 fps video with synchronized audio in one pass. Weights and a ComfyUI pack are on Hugging Face. The license is the MiniMax H3 Community License, not OSI open source: the EU, UK, South Korea and the United States are excluded territories, and commercial use above USD 20 million yearly revenue needs a separate written authorization.