comfyui
TLDR
Start ComfyUI server
SYNOPSIS
python main.py [options]
comfyui [options]
DESCRIPTION
ComfyUI is a node-based workflow system for Stable Diffusion and Flux image generation. It provides both a visual graph editor (web UI) and a REST API for programmatic use.
Workflows are built by connecting nodes: model loaders, samplers, VAE decoders, prompt encoders, and more. Complex pipelines (img2img, inpainting, ControlNet, LoRA stacking) are constructed visually without code.
The API accepts workflow JSON, enabling headless batch generation and integration with scripts. Workflows created in the web UI can be exported and run via the API.
An extensive ecosystem of custom nodes adds capabilities: video generation, face restoration, upscaling, IP-Adapter, and more. Custom nodes are installed into the `custom_nodes/` directory.
Install via pip (`pip install comfyui`) or clone the repository. Models are placed in the `models/` directory tree.
PARAMETERS
--port PORT
Server port (default: 8188).--listen ADDR
Listen address (default: 127.0.0.1, use 0.0.0.0 for network).--cpu
Run on CPU only.--cuda-device ID
CUDA GPU device index.--lowvram
Low VRAM mode for GPUs with limited memory.--dont-print-server
Suppress server output.--output-directory DIR
Custom output directory.--temp-directory DIR
Custom temp directory.--auto-launch
Auto-open browser on start.--disable-auto-launch
Prevent auto-opening browser.
CAVEATS
Web UI requires a modern browser. GPU strongly recommended. Custom nodes may conflict. Model files are large (2-10+ GB each). Python 3.10+ required. Complex workflows can consume significant VRAM.
HISTORY
ComfyUI was created by comfyanonymous in 2023 as a modular alternative to the Automatic1111 web UI. Its node-based design attracted power users who needed flexible, reproducible generation pipelines. It became one of the most popular Stable Diffusion interfaces, with a large community building custom nodes.


