Mochi 1
An open-source text-to-video model that runs locally through ComfyUI or Python and supports LoRA fine-tuning under Apache 2.0.
Mochi 1 generates video from written prompts for creators and developers who want to run a video model on their own hardware or adapt it to their own footage.
- Photorealistic video: Generate 480p clips from text prompts with a focus on realistic motion and following the requested scene.
- Local generation: Use ComfyUI or the Python tools, which include a Gradio interface and an API for adding video generation to other software.
- LoRA fine-tuning: Train adaptations on your own videos and use fine-tuned LoRA files in safetensors format.
The code and downloadable model weights are open source under Apache 2.0. Genmo also offers a hosted browser playground, separate from local generation.
Hardware needs depend on the implementation. The Python implementation supports multiple GPUs but needs approximately 60GB of VRAM on a single GPU; Genmo recommends an H100. ComfyUI optimizations can reduce inference memory use to below 20GB of VRAM. Fine-tuning requires an H100 or A100 with 80GB of memory.
Mochi 1 is less suited to animated styles. Extreme motion can cause warping or other distortions.
Logotyper
