Reactor Unveils Real-Time Open-Source Video AI 'FastH3-Omni' with API Preview Through October 20
Reactor has announced FastH3-Omni, a real-time open-source video generation model built on MiniMax H3 and post-trained with FastVideo and Nuva. Featuring first-
On October 14, 2026, AI video infrastructure platform Reactor (@reactorworld) officially introduced 'FastH3-Omni', an open-source real-time video generation model built upon MiniMax's realtime H3 foundation architecture. An exclusive preview is currently available through the Reactor API through October 20, 2026.
Image source: Reactor (@reactorworld)
FastH3-Omni was post-trained in partnership with FastVideo (@haoailab) and Nuva (@nuvalab). Reactor describes the model as "the fastest and highest-quality open-source version of realtime H3 by @MiniMax_AI," claiming frontier-level performance.
Post-Trained on MiniMax H3: Real-Time Open-Source Video Infrastructure
FastH3-Omni builds upon MiniMax's H3, an architecture designed for real-time video generation. Post-trained jointly with FastVideo and Nuva, FastH3-Omni is tuned for real-time interaction and API-driven video generation pipelines.
Multimodal Controls: Keyframe Conditioning, Audio Dialogue, and Reference Images
FastH3-Omni incorporates multiple granular control mechanisms for generative video workflows:
- First and Last Frame Conditioning: Allows creators to lock both the opening and closing frames of a sequence, providing control over transitions and scene continuity.
- Audio Input and Robust Dialogue Synchronization: Directly accepts voice audio inputs to generate character performances and dialogue timing aligned with speech.
- Up to 3 Reference Images: In addition to text prompts, the model accepts up to three reference visual inputs to help maintain subject and stylistic consistency across generations.
Exclusive Reactor API Preview Through October 20
FastH3-Omni is currently accessible alongside developer documentation at docs.reactor.inc.
This exclusive preview through the Reactor API is scheduled to run through October 20, 2026. Developers can test endpoints and review specifications directly via the official API overview documentation. Specific timelines for standalone weight releases or self-hosting support were not detailed in the initial announcement.
Sources
- Reactor Official Announcement on X: Introducing FastH3-Omni
- Reactor Developer Documentation: FastH3-Omni Model API Overview