Fal’s H3 Max Live enables faster-than-real-time AI video generation

Fal announced H3 Max Live, a variant of Minimax’s H3 model that has been post-trained for improved cost and quality, then optimized for Fal’s in-house inference engine.

The optimization delivers a reported 35x speed improvement over the official Minimax H3 endpoint, enabling video generation faster than real time.

Users can interact with the system via a chat interface: typing a prompt results in generated video appearing on screen within seconds.

Independent observers, including Ethan Mollick, reported that H3 Max Live can produce reasonably high-quality AI video in less time than it takes to watch it, including prompt enhancement time.

Third-party developers demonstrated the capability by building an infinite, interactive live stream where chat input dynamically generates subsequent video frames, creating an ongoing, AI-directed broadcast.

The demonstration underscores the feasibility of real-time, on-the-fly video generation, though early outputs were characterized as low-quality and chaotic, described as 'slop' with no coherent plot.

Despite the early limitations, the breakthrough signals a shift toward real-time generative video and interactive agentic media experiences.