← Back to live feed1 story
fal developed H3 Max, a post-trained variant of MiniMax's open-weight H3 model, that generates 15 seconds of video in 9 seconds. The model operates 50x faster than the original H3 while maintaining quality, enabling AI video to be produced faster than it can be viewed.
MiniMax said the achievement demonstrates the potential of open-weight models to advance latency and quality for interactive worlds and real-time storytelling. Developer levelsio used the model to create "Infinite Slop," a live stream where user chat inputs are generated as sequential video in real time.
Sign in to suggest edits
Key sources
- SOURCE@minimax_ai“Faster-than-real-time video is no longer just an idea”x.com
- SUPPORT@minimax_ai“can it generate faster than Twitch can stream?”x.com
- SUPPORT@levelsio“It generates 15 seconds of video in 9 seconds!”x.com
- SOURCE@levelsio“It generates 15 seconds of video in 9 seconds!”x.com
- SOURCE@fal“An infinite broadcast where every frame is generated on the fly and every scene is directed by chat”x.com
- SOURCE@fal“The model holds the thread across scenes instead of resetting every clip”x.com
- SUPPORT@fal“Watch and prompt it live here!”x.com
- SOURCEhuggingnewshuggingnews.com