ComfyUI - Node-Based AI Creation Engine

ComfyUI is the most powerful and modular AI engine for content creation that gives visual professionals complete control over every model, parameter, and output. Its node-based graph interface enables creatives to generate images, videos, 3D models, audio, and more without writing code.

Key features:

• Node/Graph Interface: Visual workflow builder for complex AI pipelines
• Multi-Model Support: Native support for 100+ open-source models including SD3.5, Flux 2, Wan 2.2, Hunyuan3D, MoChi, Stable Audio
• Smart Execution: Only re-executes changed parts of workflows between generations
• Memory-Efficient: Smart offloading allows running large models on GPUs with as low as 1GB VRAM
• Offline-First: Works fully offline; core never downloads anything unless configured
• Production-Ready: API endpoints for integration into production pipelines
• App Mode: Complex workflows exposed through simple UIs

ComfyUI is built on Python with a flexible backend using PyTorch and a modern Vue.js frontend.

Core Backend:

• Python 3.12+ (3.13/3.14 supported)
• PyTorch 2.4+ with CUDA/ROCm/XPU support
• aiohttp for async server
• SQLAlchemy for database operations
• Pillow for image processing
• numpy, scipy for computations

Frontend:

• Vue.js 3 (separate repo: ComfyUI_frontend)
• TypeScript
• Tailwind CSS
• Hosted via comfyui-frontend-package PyPI

Supported Hardware:

• NVIDIA GPUs (CUDA 12.6/13.0)
• AMD GPUs (ROCm 7.2, DirectML)
• Intel Arc GPUs (torch XPU)
• Apple Silicon (M1/M2/M3 via Metal)
• Ascend NPUs, Cambricon MLUs
• CPU-only mode (slow)

Optional Dependencies:

• transformers, tokenizers (LLM support)
• sentencepiece (tokenizer)
• safetensors (safe model loading)
• kornia (computer vision)
• spandrel (model loading)
• comfy-kitchen (comfyorg integration)
• av (video encoding)
• glslang, PyOpenGL, glfw (3D rendering)

Core Requirements:
- comfyui-frontend-package
- comfyui-workflow-templates
- comfyui-embedded-docs
- torch, torchvision, torchaudio
- numpy>=1.25.0
- einops
- transformers>=4.50.3
- tokenizers>=0.13.3
- sentencepiece
- safetensors>=0.4.2
- aiohttp>=3.11.8
- pyyaml, Pillow, scipy, tqdm
- psutil, alembic, SQLAlchemy>=2.0.0
- filelock, av>=14.2.0
- comfy-kitchen>=0.2.8
- requests, simpleeval>=1.0.0
- blake3, kornia>=0.7.1, spandrel

The easiest way to get started with ComfyUI is through the official Desktop Application, available for Windows and macOS.

Advantages:

• One-click installation
• Built-in model management
• Auto-updates
• No command-line knowledge required

Steps:

1. Visit https://www.comfy.org/download
2. Download the installer for your OS
3. Run the installer
4. Launch ComfyUI

# Download from official website
# Windows/macOS: https://www.comfy.org/download

# Alternative: Use comfy-cli
pip install comfy-cli
comfy install
comfy launch

The portable package provides immediate access to ComfyUI with no installation needed. Simply extract and run.

Requirements:

• Windows 10/11
• NVIDIA GPU (20 series or newer)
• 7-Zip for extraction

# Download portable package
# https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia.7z

# Extract with 7-Zip
7z x ComfyUI_windows_portable_nvidia.7z

# Run (NVIDIA GPU)
run_nvidia_gpu.bat

Prerequisites:

• Python 3.12+ (3.13 recommended)
• PyTorch with appropriate backend
• Git (for cloning repo)

# 1. Clone repository
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI

# 2. Install PyTorch for your GPU

# NVIDIA (CUDA 13.0)
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130

# AMD (ROCm 7.2, Linux)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2

# Apple Silicon
pip install torch torchvision torchaudio

# 3. Install ComfyUI dependencies
pip install -r requirements.txt

# 4. Run ComfyUI
python main.py

# Access at http://127.0.0.1:8188

After installation, run ComfyUI with python main.py.

Common Flags:

• --port <port>: Set port (default: 8188)
• --external: Listen on all network interfaces
• --preview-method <method>: Set preview method (auto, taesd)
• --output-dir <dir>: Set output directory
• --cpu: Run on CPU only (slow)
• --enable-manager: Enable ComfyUI-Manager
• --disable-api-nodes: Disable external API nodes

# Basic run
python main.py

# Listen on all interfaces
python main.py --external

# Custom port and high-quality previews
python main.py --port 8080 --preview-method taesd

# With Manager extension
python main.py --enable-manager

# CPU only (for testing)
python main.py --cpu

AMD cards not officially supported by ROCm can be enabled via environment variables:

GFX Override:

• RX 6700, 6600 (RDNA2): HSA_OVERRIDE_GFX_VERSION=10.3.0
• RX 7600 (RDNA3): HSA_OVERRIDE_GFX_VERSION=11.0.0

# RDNA2 cards (6700, 6600)
HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py

# RDNA3 cards (7600)
HSA_OVERRIDE_GFX_VERSION=11.0.0 python main.py

# Experimental memory efficient attention
TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1 python main.py --use-pytorch-cross-attention

Master these essential shortcuts:

• Ctrl+Enter: Queue generation
• Ctrl+Shift+Enter: Queue as first
• Ctrl+Z/Y: Undo/Redo
• Ctrl+S: Save workflow
• Ctrl+O: Load workflow
• Delete: Delete node
• Space+Drag: Pan canvas
• P: Pin node
• Ctrl+G: Group nodes
• F: Toggle menu

Essential Shortcuts:
Ctrl+Enter - Queue generation
Ctrl+Shift+Enter - Queue as first
Ctrl+Z/Y - Undo/Redo
Ctrl+S - Save workflow
Ctrl+O - Load workflow
Ctrl+G - Group nodes
Delete - Delete node
Space+Drag - Pan canvas
P - Pin node
F - Toggle menu
R - Refresh graph
H - Toggle history
Q - Toggle queue

Official Resources:

• Website: https://www.comfy.org/
• Documentation: https://docs.comfy.org
• GitHub: https://github.com/comfyanonymous/ComfyUI

Community:

• Discord: https://discord.com/invite/comfyorg
• Matrix: #comfyui_space:matrix.org
• Examples: https://comfyanonymous.github.io/ComfyUI_examples/

Official Links:
Website: https://www.comfy.org/
Docs: https://docs.comfy.org
GitHub: https://github.com/comfyanonymous/ComfyUI
Discord: https://discord.com/invite/comfyorg
Matrix: #comfyui_space:matrix.org
Examples: https://comfyanonymous.github.io/ComfyUI_examples/