Installation and First Steps
Hardware Requirements
Recommended GPU
| Configuration | VRAM | Supported Models |
|---|---|---|
| Minimum | 4 GB | SD 1.5 (slow) |
| Recommended | 8 GB | SD 1.5, SDXL |
| Comfortable | 12 GB+ | SD 1.5, SDXL, Flux |
| Optimal | 24 GB+ | All models, high resolution |
GPU Support
- NVIDIA (CUDA): native support, best performance, recommended
- AMD (ROCm): supported on Linux, decent performance
- Apple Silicon (MPS): supported, good performance on M1/M2/M3
Installation
Method 1: Manual Installation (recommended)
# Install Python 3.10 or 3.11
python --version
# Install PyTorch (NVIDIA CUDA 12.1)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
# Clone and setup ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
# Launch
python main.py
The interface will be available at http://127.0.0.1:8188.
Method 2: Portable Package (Windows)
Download from the official GitHub Releases page, extract, and run run_nvidia_gpu.bat.
Method 3: Docker
docker run -it --gpus all -p 8188:8188 \
-v $(pwd)/models:/app/models \
-v $(pwd)/output:/app/output \
comfyanonymous/comfyui
File Organization
ComfyUI/
├── models/
│ ├── checkpoints/ # Main models (.safetensors)
│ ├── loras/ # LoRAs (style/subject adapters)
│ ├── controlnet/ # ControlNet models
│ ├── vae/ # Custom VAEs
│ ├── upscale_models/ # Upscale models (ESRGAN, etc.)
│ ├── clip_vision/ # CLIP Vision models (for IP-Adapter)
│ └── embeddings/ # Textual inversions
├── input/ # Input images
├── output/ # Generated images
└── custom_nodes/ # Installed extensions
Your First Generation
- Select your model in the Load Checkpoint node
- Write a positive prompt and a negative prompt
- Configure the KSampler (25 steps, cfg 7.0, dpmpp_2m sampler)
- Set resolution in Empty Latent Image (512x512 for SD 1.5, 1024x1024 for SDXL)
- Click Queue Prompt
Installing ComfyUI Manager
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git
Restart ComfyUI. A new Manager button appears in the menu, allowing you to install custom nodes, missing nodes, and models directly from the interface.
Saving and Loading Workflows
- Ctrl + S: save as JSON
- Ctrl + O: open a JSON file
- Drag and drop a JSON or PNG image onto the canvas to load a workflow
- Workflow metadata is embedded in every generated PNG
Common Troubleshooting
- CUDA out of memory: reduce resolution, use
--lowvramflag - Red/missing nodes: install via ComfyUI Manager → Install Missing Nodes
- Black or noisy image: check VAE connection, increase steps, lower cfg