ComfyUI
by comfyanonymous and contributors; organisation Comfy Org
Open-source node-based interface and inference engine for image, video, audio and 3D generation, runnable fully offline.
See https://github.com/comfyanonymous/ComfyUI
Facts (2026-09-30)
- Licence: GPL-3.0 (the earlier note said MIT, which was wrong).
- GitHub: about 135.6k stars, 16.1k forks.
- Platforms: Windows, macOS, Linux; runs on consumer GPUs (README cites as low as 4 GB VRAM) with asynchronous weight streaming and quantised models.
- Offerings in docs: local install, Comfy Cloud (hosted GPUs), Comfy Router (unified API endpoint) and a CLI.
Supported models (README highlights)
- Image: Stable Diffusion family, SDXL, SD3.5, Flux.1/2, Qwen Image, Hunyuan Image 2.1, Ideogram 4.
- Video: Wan 2.1/2.2, LTX-Video, HunyuanVideo, CogVideoX.
- Audio: ACE-Step 1.5, Stable Audio 3, MiniMax Music.
- 3D/vision: Hunyuan3D 2.1, SAM 3/3.1. Text: Gemma 3/4, Qwen3.
Superpowers
Composable graphs: chain generation, inpainting, ControlNet/pose conditioning, LoRA, upscaling and detailers in one reusable workflow JSON that can be shared. Good for private, offline and reproducible pipelines; steeper learning curve than hosted tools.
Notes
Open-weight models only run locally; closed models (e.g. Wan 3.0 API, Seedream) need partner/API nodes or cloud hosts. Pricing of Comfy Cloud: unverified.
Related
flux-dot-1, wan-2-dot-5, meta-sam-3, fal-dot-ai
Sources
- https://github.com/comfyanonymous/ComfyUI (accessed 2026-09-30)
- https://docs.comfy.org/ (accessed 2026-09-30)