Stable Diffusion Intro: Guide to Installing with ComfyUI

Assuming I will write articles sharing results of playing with Stable Diffusion from now on, writing introduction method every time is tedious, so I will write an article as reference. Although it is just explaining content written in repository by chewing it down, those who want to read by themselves can go read from link ↓.

GitHub - Comfy-Org/ComfyUI: The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. - Comfy-Org/ComfyUI

github.com

Introduction Method

In this chapter I introduce steps to install main body. If you introduce extension ComfyUI-Manager together, introduction of other extensions and models becomes easy on UI. Since it is quite convenient, I took liberty to include it in steps. (Those not needing it please skip installing step)

Method 1: Using Standalone version on Windows (NVIDIA GPU, or CPU execution only supported)

1. There is a link Direct link to download right below link below so download from there and unzip with 7zip.
   • 
     GitHub - Comfy-Org/ComfyUI: The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.

     The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. - Comfy-Org/ComfyUI

     github.com
2. Install ComfyUI-Manager
   • Hit following command inside Unzipped Directory/ComfyUI/custom_nodes directory
   • git clone https://github.com/ltdrdata/ComfyUI-Manager.git
3. Execute run_nvidia_gpu.bat (In case of NVIDIA GPU), or run_cpu.bat (In case of CPU).
4. Complete (Web application launches automatically)

Method 2: Method using comfy-cli (NVIDIA, AMD, Intel GPU seem supported)

Those who don’t want to pollute local please use disposable python environment like venv.

GitHub - Comfy-Org/comfy-cli: Command Line Interface for Managing ComfyUI

Command Line Interface for Managing ComfyUI. Contribute to Comfy-Org/comfy-cli development by creating an account on GitHub.

github.com

1. pip install comfy-cli
2. comfy install ComfyUI-Manager is also installed together
3. comfy launch
4. Access displayed URL from browser

Method 3: Manual Install

In case of NVIDIA, AMD

1. git clone to arbitrary environment
   • git clone https://github.com/comfyanonymous/ComfyUI.git
2. Install GPU specific Dependency (cuda etc.)
   • AMD
     • pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0
   • NVIDIA
     • pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121
3. Install other Dependencies
   • pip install -r requirements.txt
4. Install ComfyUI-Manager
   • Hit following command inside custom_nodes directory
   • git clone https://github.com/ltdrdata/ComfyUI-Manager.git
5. python main.py
6. Access displayed URL from browser

In case using other than NVIDIA, AMD

Intel GPU, Apple Mac silicon, DirectML can also be handled with Manual Install. Maybe because they are minor, steps are slightly different each, so refer to repository and try working if curious.

Method 4: Method to install via StabilityMatrix

There is a tool called StabilityMatrix that manages Stable Diffusion related tools collectively. Please refer to following article for installation method.

StabilityMatrixの導入方法：Stable Diffusion関連ツールを効率的に管理

>-

blog.otama-playground.com

Conclusion

If application can be displayed on browser, for now intro edition is over.