Stable Diffusion WebUI (Automatic1111) Setup Guide

This guide provides instructions on how to set up and host your own image generation model using Automatic1111's Stable Diffusion WebUI.

Prerequisites/Assumptions I Will Need To Make

• Operating System: Windows/Linux/Mac
• GPU: Either your system has a dedicated NVIDIA GPU with at least 8GB VRAM (higher VRAM recommended for larger models) OR you know how to get access to a virtual machine with similar specs (either through Google Cloud or runpod)
  • The A1111's repo has installation guide if you have an AMD or Intel GPU, but I have not tested those steps yet. We will be using the instructions for NVIDIA GPUs for this workshop.
• Drivers: Ensure you have the latest NVIDIA drivers installed
• Some Other Requirements:
  • Python 3.10.6 (if you have an older or newer Python version, you will need to uninstall and then reinstall this version)
  • Git
  • CUDA, since we will be focusing on NIVDIA GPUs

Installation

This is how I do it on both Windows and Linux. The A1111's repo has very detailed instructions on how to install on different operating systems.

1. Clone the Repository

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

2. Navigate into the Directory

cd stable-diffusion-webui

2. Get Some Models

• Download a Stable Diffusion model (e.g., 3.0, SDXL) from huggingface.
• Place the .safetensors file in the models/Stable-diffusion directory.
• Experiment with different model sizes to figure out what works best for your setup!

Running the WebUI

To start the server, navigate into the directory, then use:

./webui.sh --listen  # For Linux/Mac
webui-user.bat       # For Windows

• By default, the WebUI runs on http://localhost:7860

• To make it accessible over your network, add the --listen flag:

  ./webui.sh --listen

• You can also configure other parameters inn the webui-user.bat file for Windows, or webui-user.sh file for Linux.

• My prefered commandline arguments for NVIDIA GPUs:

   --device-id=0 --port [whatever OPEN port number you prefer] --no-half-vae --xformers --listen --api

Generate Some Images!

• Generate with whatever prompts you would like!
• Experiment with the different parameters in the UI!

Advanced: Use the API in your own code

• With the --api tag, you can read the API documentations at the /docs subdirectory (i.e. localhost:7860/docs); or you can also click the API link at the bottom of the web UI.
• In your scripts, send requests to the /sdapi/v1/txt2img endpoint.
• For example, if you web UI is hosted at the default port 7860, you would point your requests to http://localhost:7860/sdapi/v1/txt2img with the appropriate json payload for parameters.
• See generate.py for an example (thanks Dr. Maiti for letting me steal your code)
• Before running the script, make sure you have:
  • A working Python environment (which you should already have if you are able to run the web ui)
  • The following Python packages:
    • requests
    • Pillow
• You can install these packages using pip

   pip install requests Pillow

---

References

• Automatic1111 WebUI GitHub
• Stable Diffusion Models