Image & Vision MIT

ComfyUI

Node-graph Stable Diffusion UI for complex workflows, video generation, and reproducible image pipelines. The most extensible and powerful local image gen UI in 2026.

Website
Platforms: windowsmacoslinux

ComfyUI is a node-based interface for Stable Diffusion and friends. Instead of a single form with sliders, every step — loading a checkpoint, encoding a prompt, sampling, VAE-decoding, upscaling — is a node in a graph that you wire together. That sounds forbidding, but the payoff is that a graph is reusable, shareable, and composable in ways a fixed UI can never be. When someone posts a workflow that does SDXL + ControlNet + AnimateDiff + a custom upscaler, you download one JSON file and load it verbatim. By 2026 it has become the de facto backend for most serious local image work; SwarmUI, Frontpage, and several hosted products all run ComfyUI under the hood.

Key Features

  • Graph-based, reproducible workflows. A .json file fully describes the pipeline — model, sampler, seeds, every parameter. Re-running it produces identical output on the same hardware, and sharing a workflow means sharing the graph, not a screenshot of sliders.
  • Frontend-agnostic backend. ComfyUI exposes an HTTP and WebSocket API that other UIs drive directly. SwarmUI ships ComfyUI as its engine; many “one-click” installers bundle it unseen. You can learn the graph once and keep using it from whichever frontend you prefer.
  • Largest extension ecosystem. Hundreds of custom nodes cover ControlNet, IPAdapter, AnimateDiff, video (WAN, Hunyuan Video, CogVideoX), Flux, SD3.5, and upscaling. The ComfyUI Registry (comfy-cli install <node>) handles dependencies and versioning.
  • First-class video support. Temporal models, frame interpolation, and batched sampling are nodes, not hacks. Video pipelines that take 30 clicks elsewhere are a single loaded graph here.

When to Use ComfyUI

Reach for ComfyUI when you outgrow a fixed UI — when you need ControlNet stacks, multi-model routing, video, or any pipeline you want to reproduce and share. It is also the right choice if you are building a product on top of local image generation: the API is stable, the graph format is portable, and you can lock a workflow to a specific ComfyUI version. Avoid it only if you want the simplest possible “type prompt, get image” experience — that is what Forge or SwarmUI are for.

Setup

# Clone and run (Python 3.10+ and a CUDA/ROCm/Metal GPU recommended)
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
pip install -r requirements.txt
python main.py            # opens http://127.0.0.1:8188

# Or use the standalone Windows release (bundled Python + torch):
# https://github.com/comfyanonymous/ComfyUI/releases

# Manager + CLI for installing custom nodes:
pip install comfy-cli
comfy node install ComfyUI-Manager

Models go in models/checkpoints (or models/diffusion_models for Flux). The Manager adds a browser-based node installer accessible from the graph canvas.

See also