First open-source autoregressive video generation model. Turn images into infinite videos with precise timeline control and natural motion quality.
Infinite Extension
Generate long videos seamlessly with no visible cuts
Timeline Control
Fine-tune every second for precise timing needs
Natural Motion
Overcome stiff AI motions with dynamic visuals
Open-Source
All models and technical reports publicly available
MAGI-1 combines autoregressive video block generation with advanced temporal modeling. It offers 24B and 4.5B parameter versions, producing high-quality motion with smooth scene transitions. Includes 61-page technical report and full open codebase.
Supports 1-10 second videos with infinite extension. Available on GitHub and Hugging Face
Content creation from images for social media
Research with open-source video generation
Video production extending existing footage
The first autoregressive video generation model supporting infinite length videos
Currently 1-10 seconds with ability to infinitely extend
Publicly available on GitHub and Hugging Face
