Portable Package

The portable package is a pre-built, self-contained version of ComfyUI. Download, extract, and run — no Python installation or system changes required.

This method supports NVIDIA, AMD, and CPU modes. See GPU Compatibility for details on each.

Step 1: Download

Go to the ComfyUI GitHub releases page and download the portable package:

• ComfyUI_windows_portable_nvidia.7z — For NVIDIA GPUs (CUDA 13.0, Python 3.13)
• ComfyUI_windows_portable_nvidia_cu126.7z — For older NVIDIA GPUs (CUDA 12.6)

The file is around 1-2 GB.

Step 2: Extract

Use 7-Zip

You must use 7-Zip to extract the archive. The Windows built-in zip extractor may fail due to long file paths.

1. Right-click the downloaded .7z file
2. Select 7-Zip → Extract to "ComfyUI_windows_portable/"
3. Wait for extraction to complete (may take a few minutes)

If extraction fails: Right-click the .7z file → Properties → check Unblock at the bottom → click Apply, then try again.

Path rules (important!):

• Choose a location with at least 50 GB of free space
• Avoid spaces and special characters in the path (e.g., D:\ComfyUI is good, D:\My Programs\ComfyUI! is bad)
• Do NOT extract to system folders like C:\Program Files, C:\Windows, or C:\ root
• Do NOT run as Administrator — this can cause permission issues with Python packages
• Keep the path short to avoid Windows 260-character path limit (see Long Paths fix)

Step 3: Download a Model

Before you can generate images, you need at least one checkpoint model. Download one and place it in the correct folder:

ComfyUI_windows_portable/
└── ComfyUI/
    └── models/
        └── checkpoints/    ← Put your .safetensors file here

Popular starter models:

• SD 1.5 (~2 GB) — fast, low VRAM, huge community
• SDXL (~7 GB) — higher quality, needs 8 GB+ VRAM
• Flux Schnell (~12 GB) — latest generation, needs 12 GB+ VRAM

See Download Models for download links and details.

Step 4: Run

Double-click the appropriate batch file in the root folder:

File	When to Use
run_nvidia_gpu.bat	NVIDIA GPU (most users)
run_cpu.bat	No dedicated GPU, or AMD GPU with DirectML

A console window will open showing startup logs. Once you see:

Starting server
To see the GUI go to: http://127.0.0.1:8188

Open http://127.0.0.1:8188 in your browser (Chrome recommended).

Updating

To update ComfyUI:

1. Navigate to the update/ folder inside the portable directory
2. Double-click update_comfyui.bat
3. Wait for the update to complete
4. Restart ComfyUI

To update Python packages (PyTorch, etc.), use the batch files in the update/ folder.

Folder Structure

ComfyUI_windows_portable/
├── python_embeded/          # Bundled Python (do NOT modify)
├── update/                  # Update scripts
│   ├── update_comfyui.bat
│   └── update_comfyui_stable.bat
├── ComfyUI/
│   ├── models/              # AI models
│   │   ├── checkpoints/     # Main models (.safetensors)
│   │   ├── loras/           # LoRA files
│   │   ├── vae/             # VAE files
│   │   ├── controlnet/      # ControlNet models
│   │   ├── upscale_models/  # Upscaler models
│   │   └── clip/            # CLIP models
│   ├── custom_nodes/        # Community plugins
│   ├── input/               # Input images for img2img
│   ├── output/              # Generated images
│   └── extra_model_paths.yaml  # Share models with other tools
├── run_nvidia_gpu.bat
└── run_cpu.bat

Sharing Models with A1111 / Forge

If you already have models from other AI tools, you don't need to download them again.

Method 1: extra_model_paths.yaml

Edit ComfyUI/extra_model_paths.yaml to point to your existing model folders:

a111:
  base_path: D:/stable-diffusion-webui/  # Your A1111 path

  checkpoints: models/Stable-diffusion
  vae: models/VAE
  loras: models/Lora
  upscale_models: models/ESRGAN
  controlnet: models/ControlNet

Restart ComfyUI after saving.

Method 2: Symbolic Links (Symlinks)

If you want ComfyUI to use an existing model folder directly, you can create a Windows symbolic link. This is more efficient than copying files:

mklink /D "ComfyUI\models\checkpoints" "D:\my-models\checkpoints"

This makes ComfyUI see models in D:\my-models\checkpoints as if they were in its own folder, without duplicating files.

Note

Creating symlinks on Windows may require Developer Mode enabled, or running the command prompt as Administrator (just for the mklink command, not for running ComfyUI itself).

Troubleshooting

Problem	Solution
Browser shows title but no UI	Update to the latest version of Chrome
CUDA out of memory error	Close other GPU apps, or add --lowvram to the bat file
torch is not compiled with CUDA	You downloaded the wrong package. Use the NVIDIA version
Red nodes in workflows	Missing custom nodes. See Install Custom Nodes
Slow generation on NVIDIA GPU	Make sure you're running run_nvidia_gpu.bat, not run_cpu.bat

Adding Launch Arguments

To add flags like --lowvram, edit the .bat file with a text editor. Find the line with main.py and add your flags:

.\python_embeded\python.exe -s ComfyUI\main.py --lowvram

Common flags:

Flag	Effect
--lowvram	Use less GPU memory (slower)
--cpu	Run entirely on CPU
--listen	Allow access from other devices on your network
--port 8189	Change the web server port
--highvram	Keep models in VRAM (faster if you have enough)

Next Steps

• Generate your first image
• Download models
• Install custom nodes
• Troubleshoot common issues