ComfyUI Installation Tutorial: Desktop, Windows Portable, and Manual Setup

ComfyUI has several installation options. Choose one based on your computer and needs:

Method 1: ComfyUI Desktop (recommended for beginners)

1. Download the Desktop version from the ComfyUI website (comfy.org).
2. Run the installer and follow the setup wizard:
   • Choose a GPU type (NVIDIA, Apple Silicon, etc.)
   • Choose an installation location (a drive with plenty of space is recommended; your models will be stored here)
   • If you have installed ComfyUI before, you can import your existing models and settings
3. The installer automatically downloads Python and the required packages. When it's done, open the interface.

The Desktop version includes ComfyUI-Manager (for managing custom nodes) and automatic updates.

Method 2: Windows portable

1. Download the portable archive from the Releases page on ComfyUI's GitHub. Choose the version for your graphics card (primarily NVIDIA; AMD and Intel versions are also available).
2. Use 7-Zip to extract it to a location with plenty of space, such as D:\ComfyUI_windows_portable. Avoid Chinese characters and spaces in the path.
3. Run one of these files in the folder:
   • run_nvidia_gpu.bat: uses an NVIDIA graphics card
   • run_cpu.bat: uses only the CPU (very slow)
4. When the terminal window shows To see the GUI go to: http://127.0.0.1:8188, open that URL in your browser.

The portable version includes its own Python installation and won't affect other programs on your computer. To update it, run the update script in the update folder.

Older NVIDIA graphics cards (such as the GTX 10 series) may require a version that is officially marked as supporting older graphics cards.

Method 3: Manual installation

This option is suitable for Linux or anyone who wants to manage the environment themselves. Install Git and Python first (use the versions recommended in the official README; currently, Python 3.12–3.13).

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
python -m venv venv
source venv/bin/activate          # Windows: venv\Scripts\activate

Install PyTorch

Install the PyTorch version for your graphics card (version numbers change over time, so check the ComfyUI README or the PyTorch website):

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

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

# Intel graphics card
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/xpu

For Apple Silicon Macs, install PyTorch by following Apple's PyTorch Metal instructions.

Install dependencies and launch

pip install -r requirements.txt
python main.py

Open http://127.0.0.1:8188.

You can also use comfy-cli

pip install comfy-cli
comfy install
comfy launch

Common launch arguments

python main.py --listen 0.0.0.0 --port 8188

To add arguments to the portable version, use a text editor to modify the python main.py line in run_nvidia_gpu.bat.

When you use --listen to allow other computers to connect, ComfyUI has no username or password protection. Use it only on a trusted local network.

Check that the GPU is being used

The terminal output at startup shows the detected device, for example:

Total VRAM 12288 MB, total RAM 32671 MB
Device: cuda:0 NVIDIA GeForce RTX 4070 : cudaMallocAsync

If it shows the CPU, PyTorch's GPU version may not be installed correctly.

Update

• Desktop: Updates are shown automatically.

• Portable: Run update/update_comfyui.bat.

• Manual installation:

  git pull
  pip install -r requirements.txt

If custom nodes stop working after an update, they usually need to be updated too. See Custom nodes and Manager.

Next steps

After installation, ComfyUI still can't generate images because you don't have a model yet. Start with the interface tour, then go to model downloads and folders to download your first model.

Updated October 12, 2026