Mochi 1 (Genmo) vs Wan (Alibaba)
A spec-by-spec comparison of Mochi 1 (Genmo) and Wan (Alibaba): pricing, free tier, clip length, resolution, watermark, commercial use and API access.
| Mochi 1 (Genmo) | Wan (Alibaba) | |
|---|---|---|
| Pricing | Open source | Open source |
| Free tier | Free weights (Apache 2.0) and a free playground on genmo.ai | Free to download and run locally (Apache 2.0) |
| Max clip length | ||
| Max resolution | 480p | 720p |
| Watermark on free plan | ||
| Commercial use | Apache 2.0 license | Apache 2.0 license |
| Developer API | ||
| Open source | Yes | Yes |
Mochi 1 is an open text-to-video model from Genmo, built on its Asymmetric Diffusion Transformer architecture with about 10 billion parameters. Weights and inference code are released under Apache 2.0, and there is a LoRA fine-tuning trainer. Genmo also runs a free web playground. It is best for researchers and tinkerers who want a permissively licensed model to run or fine-tune.
- Permissive Apache 2.0 license
- Native ComfyUI support, including consumer-GPU workflows
- Includes a LoRA fine-tuning trainer for your own videos
- Reference repo needs about 60GB VRAM on a single GPU (at least one H100 recommended); ComfyUI can run it in under 20GB
- Released as a research preview at 480p only, and Genmo says it handles animated content poorly
- The only public checkpoint is still mochi-1-preview from October 2024
Wan is a family of open-weight video generation models from Alibaba. The Wan2.2 release includes 14B mixture-of-experts text-to-video and image-to-video models, a compact 5B model that does both, a speech-to-video model, and Wan-Animate for character animation and replacement. It is aimed at developers, researchers and ComfyUI/Diffusers users who want to run or fine-tune video models themselves.
- Permissive Apache 2.0 license, and the team claims no rights over generated content
- Broad task coverage: T2V, I2V, speech-to-video and character animation/replacement
- Integrated into Hugging Face Diffusers, with Hugging Face/ModelScope demo spaces
- 14B models need a GPU with at least 80GB VRAM for single-GPU inference
- 5B TI2V model needs at least 24GB VRAM (e.g. RTX 4090)
- Open models output at most 720P (480P/720P)
Specs last verified 2026-10-23 (Mochi 1 (Genmo)) and 2026-10-23 (Wan (Alibaba)). See each tool page for sources.