October 13, 2026
FastH3-Omni: video with sound from a single prompt, preview on the Reactor API until Oct 20, 2026
An open version of MiniMax's H3 now generates video and sound together in one go: the preview on the Reactor API is open until Oct 20, 2026.
MiniMax (official)
@minimax_ai
FastH3-Omni is now on Reactor: video with sound synced to the picture, from a single prompt. Now your imagination has its own sound studio.
Introducing FastH3-Omni. The fastest and highest-quality open-source version of realtime H3 by @MiniMax_AI, post-trained by FastVideo @haoailab with @nuvalab. Exclusive preview on the Reactor API until 10.20.2026. • First and last frame conditioning • Audio input + strong dialogue • Up to 3 reference images • Frontier-level performance
· 1.1K views
One prompt in. Out comes a video with the sound already laid over the picture.
Sound and picture are usually assembled in two passes by different models. FastH3-Omni delivers both layers from a single request.
Realtime H3 used to live at MiniMax as its own model. Now its open version has been fine-tuned by the FastVideo team together with nuvalab, and the result is showcased on the Reactor platform.
What it takes as input. Besides text, the model accepts audio and drives on-screen dialogue to match it. You can also set the first and last frames of a clip and attach up to three reference images.
Where to try it. Only through the Reactor API and only until Oct 20, 2026: the preview is exclusive. Reactor calls the quality frontier-level; that is the platform's own claim, not an independent measurement.
The exclusive preview on the Reactor API ends Oct 20, 2026.