Genmo Mochi Pipeline (Self-Hosted OSS)

aid: genmo-ai:genmo-ai-mochi-pipeline-api
name: Genmo Mochi Pipeline (Self-Hosted OSS)
tags:
- Video Generation
- Text-to-Video
- Open Source
- Self-Hosted
image: https://kinlane-images.s3.amazonaws.com/shared/apis-json/apis-json-logo.jpg
humanURL: https://github.com/genmoai/mochi
properties:
- url: https://github.com/genmoai/mochi
  type: Documentation
- url: https://huggingface.co/genmo/mochi-1-preview
  type: SourceCode
- url: https://www.genmo.ai/blog/mochi-1-a-new-sota-in-open-text-to-video
  type: Blog
endpointsModeled: true
description: The first-party access surface is the open-source model itself. The Mochi repo (Apache-2.0)
  ships a "simple composable API" in Python - MochiSingleGPUPipeline built from T5ModelFactory, DitModelFactory,
  and DecoderModelFactory - invoked with prompt, resolution, num_frames, inference steps, guidance scale,
  and seed to generate 480p video locally. It also provides a cli.py, a Gradio UI (gradio_ui.py), and
  native ComfyUI support. This is a code library / local inference pipeline, not a hosted network API;
  weights (~60GB VRAM) are downloaded from Hugging Face. There is no Genmo REST endpoint - self-host it
  or use a partner (see the Replicate and fal.ai entries).