Mochi 1 is built on a custom “Asymmetric Diffusion Transformer” architecture and trained from scratch by Genmo. Its permissive license makes it a common base for developers who want to self-host rather than call a hosted API.
Mochi 1
by Genmo
Open-source 10B-parameter text-to-video model released under Apache 2.0.
- 10 billion parameters — described by Genmo as the largest openly-released video model at launch
- Fully open-source under Apache 2.0, including for commercial use
- Generates 480p video, up to 5.4 seconds, at 30fps
- Free playground available at genmo.ai/play; weights on Hugging Face and GitHub