ComfyUI

ComfyUI is an open-source AI creation engine maintained by Comfy-Org (originally the comfyanonymous team), first released on GitHub in January 2023. It positions itself as "the most powerful and modular AI engine for content creation", aimed at visual professionals who demand control over every model and parameter. It abstracts inference pipelines for image, video, 3D, and audio diffusion models into a drag-and-drop node graph, letting users compose samplers, ControlNet, LoRA, IP-Adapter, and dozens of other components without writing code. Official distributions include the ComfyUI Desktop app and ComfyUI Cloud.

Core capabilities center on the node graph + multi-modal model adapter pattern. Supported model families span image generation (SD1.x, SD2.x, SDXL, SDXL Turbo, SD3, SD3.5, Stable Cascade, Pixart Alpha and Sigma, AuraFlow, HunyuanDiT, Flux and Flux 2, Lumina Image 2.0, HiDream, Qwen Image, Hunyuan Image 2.1, Z Image, Ernie Image), image editing (Omnigen 2, Flux Kontext, HiDream E1.1, Qwen Image Edit), video (Stable Video Diffusion, Mochi, LTX-Video, Hunyuan Video, Wan 2.1 and 2.2, Hunyuan Video 1.5), audio (Stable Audio, ACE Step), and 3D (Hunyuan3D 2.0). Runtime features include an asynchronous queue system (multiple workflows in flight), smart VRAM management with auto-offloading (1GB VRAM can run large models), incremental execution that only re-runs changed nodes, CPU fallback mode (--cpu, slow but available), full fine-tuning artifact support (LoRA, LoCon, LoHa, Hypernetwork, Textual Inversion), built-in TAESD latent previews, round-trip read/write of full workflows (including seeds) from PNG/WebP/FLAC metadata, JSON workflow import/export, an integrated Safety Checker and invisible-watermark, App Mode for wrapping any workflow into a simple UI, and API endpoints for production pipelines.

Use cases include: diffusion researchers running ablations, AIGC engineers wiring workflows into production pipelines, independent creators doing fine-grained reproducible control (hires fix, inpainting, area composition, ControlNet guidance), teams sharing parameters via workflow JSONs, and desktop users running SDXL / Flux end-to-end. Hardware floor: 1GB VRAM minimum (with smart offload); GTX 1080Ti runs CogVideoX-2B; RTX 3060 runs CogVideoX-5B; Apple Silicon / Linux / Windows all supported. Onboarding paths: download the desktop app (unified macOS / Windows / Linux), use the portable archive, or git clone and pip install -e .. ComfyUI-Manager is a community-maintained extension manager that installs and manages thousands of custom nodes and models from inside the UI.

Licensed under GPL-3.0: the project is free, open-source, and commercially usable, but anyone who modifies it and offers it as a service must publish the derivative source (this matches the AGPL-3.0 of contemporaneous WebUI projects). ComfyUI-Manager lives at Comfy-Org/ComfyUI-Manager under the same GPL-3.0 license. Note: the GitHub API indicates the canonical repo has moved to Comfy-Org/ComfyUI, while comfyanonymous/ComfyUI remains reachable and continues to receive issues and releases in parallel.