Faster Than Playback: MiniMax H3 Max Live Turns AI Videos into Infinite Livestreams, Real-Time Prompts Alter the Plot in Seconds

Beating AI News Flash: AI inference platform fal engineer Rehan Sheikh has directly connected H3 Max to Twitch. Since the model generates videos faster than viewers can play them, the livestream can simultaneously broadcast while continuing to generate subsequent frames in the background, theoretically enabling an endless stream. Sheikh calls it 'Infinite Cross-Dimensional Television.' MiniMax's official account also shared this experiment. fal later conducted further experiments with H3 Max Live. The content in the livestream is not pre-generated; viewers can directly input `!prompt` in the Twitch chatroom. According to fal, new requests can appear in the video within seconds, and every scene can be influenced by the chatroom. The AI has no fixed script or duration—so long as it continues generating, the program can keep evolving. The biggest obstacle to this type of format in the past was generation speed. If a video plays for 15 seconds but the next segment takes several minutes to generate, it’s impossible to achieve true livestreaming. H3 Max has reduced this waiting time to less than the playback time, allowing the AI to prepare the next segment while viewers are still watching the current one. For the first time, video models have shifted from 'waiting for output' to continuously supplying real-time content. H3 Max itself is a version trained by fal Research based on MiniMax’s open-source H3, optimized for fal's inference system to accelerate performance. Official data shows that generating a 5-second video takes less than 3 seconds, with throughput approximately 35 times higher than MiniMax’s official H3 API. Currently, on the Artificial Analysis leaderboard for video generation with sound, it ranks first with an Elo score of 1202, and third for text-to-video generation with an Elo score of 1235, trailing only Wan 3.0 and Gemini Omni Flash. [Original Link]