Genmo
Open video-generation research lab — Mochi 1's 10B-parameter open-weights model plus a hosted playground.
About
Genmo is a research lab building open video world models, with Mochi 1 as its flagship: a 10-billion-parameter text-to-video diffusion model released under Apache 2.0, trained from scratch on a novel Asymmetric Diffusion Transformer architecture. You can generate in the browser playground, download weights from Hugging Face for local runs, or build via partner APIs and ComfyUI workflows. The lab publishes architecture details and benchmarks positioning Mochi competitively on motion quality against closed generators.
The honest picture: Mochi 1 preview caps at 480p and 5.4-second clips — impressive for open weights, but far behind commercial tools on resolution and duration. Running the full model needs serious GPU muscle (originally 4x H100), so most people will use the hosted playground, which runs on credits. Training-data disclosure has been vague, which matters if you care about provenance.
What it can do
- Mochi 1: 10B-parameter open-weights text-to-video (Apache 2.0)
- Hosted browser playground with Replay and Mochi modes
- Hugging Face weights plus ComfyUI workflow support
- 30fps output with strong motion realism
- Published architecture and benchmark writeups
Setup
Try the hosted playground in the browser — no install. For local runs, pull weights from Hugging Face (genmo/mochi-1-preview) and follow the GitHub/ComfyUI instructions; expect heavy GPU requirements.
Prerequisites
A Genmo account for the playground; 4x H100-class GPUs (or cloud credits) for full local inference.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 48; Trending score 72/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.