Video Gen · Model Guide · 4 min read
Mochi 1 10B
Genmo's Apache-licensed 10B text-to-video model — smooth, coherent motion that a single 24 GB GPU can host.
| Developer | Genmo |
|---|---|
| Availability | Open weights |
| Commercial license | Apache 2.0 — unrestricted commercial use of model and outputs. One of the cleanest licenses in generative video. |
Basic Information
Mochi 1 was the open-weights video model that proved frontier-style motion coherence could ship under a permissive license. Its 10 billion-parameter asymmetric diffusion transformer generates five-second 480p clips with unusually fluid human motion and prompt adherence, and its Apache 2.0 license means anyone can run, modify, or build products on it.
It is text-to-video only (no image conditioning in the base release), and generation is slow on consumer hardware — think minutes per clip, not seconds — but it runs entirely offline with no metering.
Best use cases
- Fully private local video generation on a 24 GB GPU
- Commercial clip generation with no hosted-service terms (Apache 2.0)
- Custom motion styles via fine-tuning — the weights are yours to train
- Research and pipeline experimentation in ComfyUI/Diffusers
Computational Requirements
Single 24 GB card territory for the base release with offloading; quantized community builds reduce the bar, at the cost of speed and fidelity.
| Configuration | Memory |
|---|---|
| FP16 (reference) | ~19 GB |
| Practical target | 24 GB card for 480p; 48 GB for comfortable runs |
Commercial license
Apache 2.0 — unrestricted commercial use of model and outputs. One of the cleanest licenses in generative video.
Check whether it fits your hardware
Mochi 1 10B is in our compatibility library — open the compatibility panel and look it up with your GPU and context length.
More guides: back to Video Gen · visit the blog