ComfyUI
Officially recommendedA modular diffusion model generation engine with a node-based interface.
Download ComfyUI
64-bit portable version (.zip), extract and run
Clicking "Download" takes you to a source confirmation page first; the download starts after confirmation.
About ComfyUI
The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.
See incorrect information? Tell us
Risk notice
The Windows portable version is not digitally signed. Some systems may show an "Unknown Publisher" warning, and SmartScreen may block the first run. You will need to unblock the file in its properties. Additionally, third-party model files may be subject to different licenses; please verify compliance on your own.
Who it's for
- AI content creators who want fine-grained control over every generation parameter
- Designers and developers who need to build reusable, complex workflows
- Enthusiasts who want to try the latest open-source image, video, and 3D models
- Users who prefer offline operation and care about data privacy
- Technical users who need to integrate generation pipelines into automated workflows
Who it's not for
- New users who want one-click generation without needing to understand parameters
- Users who only use non-Windows systems and require an official portable package
- Users who do not want to manually download and configure models
- Those who find a node-based interface difficult to use
Install guide
Windows Portable Installation
- Choose the corresponding
.7zfile based on your GPU type:
- NVIDIA GPU (20 series and newer):
ComfyUI_windows_portable_nvidia.7z - NVIDIA GPU (10 series and older):
ComfyUI_windows_portable_nvidia_cu126.7z - AMD GPU:
ComfyUI_windows_portable_amd.7z - Intel GPU:
ComfyUI_windows_portable_intel.7z
- Use 7-Zip to extract to any directory (e.g.,
D:\ComfyUI). If extraction fails, right-click the archive → Properties → check "Unblock" and retry.
- Enter the extracted
ComfyUIfolder and run the startup script according to the instructions inside (the portable package includes Python and PyTorch, so no separate installation is needed).
- Place model files (
.ckpt/.safetensors) intoComfyUI\models\checkpoints; place other models into the corresponding subfolders underComfyUI\modelsas described in the README.
- After startup, the terminal will display a local address; open
http://127.0.0.1:8188in your browser.
Common issues:
- If the NVIDIA version fails to start, update your GPU driver.
- If you are unsure about your GPU model, check it in Task Manager → Performance → GPU.
- To share models from other UIs, rename
extra_model_paths.yaml.exampleand edit it.
FAQ
Is ComfyUI free?
Yes. ComfyUI is open source under the GPL-3.0 license and free to use. The official paid cloud service, Comfy Cloud, is separate, but the local version is free.
Which operating systems are supported?
The official desktop app supports Windows and macOS; manual installation supports Windows, Linux, and macOS. This site currently provides the Windows portable installer.
How do I update ComfyUI?
The official team releases a new version roughly every week. You can download the latest portable package from the GitHub Releases page and replace the old version, or use tools like comfy-cli to assist with updating. Be sure to back up your models and workflow files.
I don't have an NVIDIA GPU. Can I still use it?
Yes. The official offering includes portable packages for AMD and Intel GPUs, and CPU-only mode is also supported (slower). Please select the corresponding version when downloading.
Where do I place model files?
Put model checkpoint files in the `models\checkpoints` folder under the ComfyUI installation directory; place VAE, LoRA, etc. in the corresponding subfolders. You can customize search paths via the `extra_model_paths.yaml` configuration file.