FastH3-Omni brings synchronized audio-video generation to Reactor preview
Reactor introduced FastH3-Omni in a post shared by MiniMax on X, describing it as an open-source version of MiniMax's realtime H3 post-trained by FastVideo with haoailab and nuvalab that generates video with synchronized audio from a single prompt. The model supports first- and last-frame conditioning and audio input, and is available as an exclusive preview on the Reactor API until Oct. 20, 2026. Reactor also calls it the "fastest and highest-quality" open-source realtime H3 version, but no independent benchmarks were supplied to support that claim.
Reactor is positioning FastH3-Omni as an open-source realtime H3 variant that produces synchronized audio and video from a single prompt, offered through a time-limited exclusive API preview.
FastH3-Omni on Reactor - video with synchronized audio, from a single prompt. Your daydreams have a sound department now.
reactor: 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