Mochi 1
Genmo
Genmo's open text-to-video model, distributed as the mochi-1-preview checkpoint under Apache 2.0. It is built on a 10B-parameter Asymmetric Diffusion Transformer (AsymmDiT) whose visual stream carries nearly four times the parameters of its text stream, paired with a 362M-parameter AsymmVAE that compresses video 128x (8x8 spatial, 6x temporal) into a 12-channel latent space. Prompts are encoded by a single T5-XXL model. Genmo calls it a research preview that generates 480p video today, and publishes weights on Hugging Face, as a direct download and as a magnet link, alongside a free hosted playground.
Strengths
- Apache 2.0 licence, one of the few genuinely permissive open video generation models
- 10B-parameter AsymmDiT, which Genmo calls the largest video generative model openly released
- AsymmVAE compresses video 128x (8x8 spatial, 6x temporal) into a 12-channel latent space
- Encodes prompts with a single T5-XXL model rather than a stack of text encoders
- Ships an inference harness including an efficient context-parallel implementation for multi-GPU runs
- Multiple distribution routes: Hugging Face, direct download, magnet link, Diffusers and ComfyUI
Caveats
- Genmo's own limitations: "The initial release generates videos at 480p today" and "In some edge cases with extreme motion, minor warping and distortions can also occur"
- "Mochi 1 is also optimized for photorealistic styles so does not perform well with animated content"
- Heavy to run: the reference repository "requires approximately 60GB VRAM when running on a single GPU" and Genmo recommends at least one H100; the Diffusers path quotes 42 GB, or 22 GB for the bf16 variant with "a slight drop in the quality"
- Text-to-video only - no image conditioning, no audio, and the checkpoint is labelled a research preview and "a living and evolving checkpoint"
- No numeric evaluation is published: the card claims prompt adherence only "in preliminary evaluation", and on safety says the models "inherently reflect the biases and preconceptions found in their training data" with organisations told to add their own protocols
- Vendor
- Genmo
- License
- Apache-2.0
- Modalities
- video
Capabilities
Resources
Pricing
Apache 2.0 weights, free to download and self-host from Hugging Face, a direct download at weights.genmo.dev or a magnet link. Genmo also offers a hosted playground: the card says "Try this model for free on our playground."