Repolode
Submit
genmoai

genmoai/mochi

Mochi 1 is an open-weight 10B text-to-video diffusion model from Genmo, released under Apache 2.0.

Open-source alternative to:RunwaySora

View on GitHub
Stars
3,762
Forks
497
Open issues
61
Last commit
16 days ago

What is it?

Mochi 1 is a text-to-video generation model released by Genmo as a research preview. It uses a 10 billion parameter diffusion transformer, called AsymmDiT, that encodes prompts with a single T5-XXL language model and pairs them with an open AsymmVAE for video compression. The model is aimed at photorealistic motion and prompt adherence, and the initial release generates 480p video.

The repository provides a Gradio web UI and a command-line script for generating clips from local weights, with a CPU offload option to reduce GPU memory use. A Python API exposes the pipeline for programmatic use. LoRA fine-tuning is supported, and the trainer is designed to run on a single H100 or A100 80GB GPU. Single-GPU inference needs roughly 60GB of VRAM, while ComfyUI integrations can run it on less memory.

The weights are available from Hugging Face and a direct download. The project notes that it is optimized for photorealistic content and performs poorly on animated styles, and that deployment in commercial products requires additional safety measures.

Key features

  • 01Generates 480p video from text prompts with a 10B-parameter diffusion model.
  • 02Provides a Gradio web UI and a CLI for local generation.
  • 03Offers a Python API for calling the pipeline in code.
  • 04Supports CPU offload and LoRA fine-tuning on a single high-memory GPU.
  • 05Ships an open AsymmVAE for compressing and decoding video latents.

Use cases

  • Researching open text-to-video generation and diffusion architectures.
  • Generating photorealistic short clips from prompts on a local GPU.
  • Fine-tuning a LoRA to adapt the model to a specific visual style or subject.
  • Building video generation features into a custom application through the Python API.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/genmoai/mochi
  2. 2

    Enter the project directory

    cd mochi
  3. 3

    Install uv and create a Python 3.11 virtual environment

    pip install uv && uv venv --python 3.11 .venv
  4. 4

    Activate the environment and sync dependencies

    source .venv/bin/activate && uv sync --locked

Things to know

  • Requires roughly 60GB VRAM on a single GPU, with at least one H100 recommended.
  • The initial release generates videos at 480p only.
  • Optimized for photorealistic styles and performs poorly on animated content.
  • Model weights must be downloaded separately and FFMPEG installed.

In collections

AI video generationAI video & audio tools

Topics

#video-generation#text-to-video#diffusion-model#generative-ai#lora#pytorch

FAQ

What is mochi?
Mochi 1 is an open-weight 10B text-to-video diffusion model from Genmo, released under Apache 2.0. Mochi 1 is a text-to-video generation model released by Genmo as a research preview. It uses a 10 billion parameter diffusion transformer, called AsymmDiT, that encodes prompts with a single T5-XXL language model and pairs them with an open AsymmVAE for video compression. The model is aimed at photorealistic motion and prompt adherence, and the initial release generates 480p video.
Is mochi free to use? What license does it have?
mochi is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is mochi still maintained?
Actively maintained. The last commit to mochi was on Oct 22, 2026.
What can I use mochi for?
Researching open text-to-video generation and diffusion architectures. Generating photorealistic short clips from prompts on a local GPU. Fine-tuning a LoRA to adapt the model to a specific visual style or subject. Building video generation features into a custom application through the Python API.
How do I install mochi?
Clone the repository: git clone https://github.com/genmoai/mochi Enter the project directory: cd mochi Install uv and create a Python 3.11 virtual environment: pip install uv && uv venv --python 3.11 .venv Activate the environment and sync dependencies: source .venv/bin/activate && uv sync --locked
Where can I get mochi?
The source code and installation instructions are on GitHub: https://github.com/genmoai/mochi

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects