* This blog post is a summary of this video.

Installing Automatic1111 Web UI with SDXL for AI Image Generation

Author: Incite AITime: 2024-03-23 03:25:00

Table of Contents

Downloading and Installing Prerequisites for Automatic1111 and SDXL

There are a few prerequisites we need to install before we can use Automatic1111 and SDXL to generate AI images. We'll be following the recommended methods listed on the Automatic1111 GitHub page to ensure proper installation and compatibility.

Python 3.10.6

Automatic1111 requires Python 3.10.6 specifically. To get started, go to python.org, click on the Windows installer, and launch the EXE file once downloaded. Make absolutely sure to check the box to add Python to your system path, as this is crucial for Automatic1111 to function properly later on. Accept all other default settings and let the install complete. If you have any other versions of Python on your system already, it's best to uninstall those first to prevent potential conflicts.

Git Version Control

We also need Git, which is a version control system that will allow us to download Automatic1111 code from GitHub, as well as update it easily later on. Go to git-scm.com and download the Windows standalone installer, accepting all default settings on the way through. Once installed, create a folder called 'a11' somewhere on your system - this will be our main Automatic1111 folder. Next, go to the Automatic1111 GitHub page and click on the green 'Code' button, then copy the HTTPS URL. Go back to your a11 folder, type 'CMD' into the location bar to open a command prompt, and type 'git clone ' to clone the code onto your local system.

Getting the Latest Automatic1111 Code

Now that we have Python and Git installed, we can download the latest Automatic1111 code. This will provide us with the core user interface and generation functionality that we'll build on later.

As outlined earlier, open up a command prompt in your empty a11 folder and type 'git clone ' to download the code. Once finished, you can update to the newest version at any time by navigating into the /stablediffusion-web-ui folder and running 'git pull'.

Downloading the SDXL Model for Enhanced Image Generation

The open-source Automatic1111 software gives us a robust platform for AI image generation, but we need deep learning models to actually generate images. We'll be using SDXL, a cutting-edge diffusion model fine-tuned specifically for extra quality and detail compared to previous models.

Go to huggingface.co and find the SDXL models for download. Grab the SDXL Base 1.0 Save Tensors model and SDXL Refiner model (links in video description) - you'll notice these are quite large at around 6GB each, so make sure you have sufficient disk space on your target drive. Once downloaded, copy these into /stablediffusion-web-ui/models/stable-diffusion/sdxl to keep them organized.

Configuring Automatic1111 for Best SDXL Performance

We're ready to launch Automatic1111 now, but we should first configure a few settings that will ensure maximum performance and quality when leveraging the SDXL model.

Right click on the web-ui-user.cmd file in /stablediffusion-web-ui and choose Edit. This will open the file settings in Notepad. Scroll down to the command line arguments section and add the following:

  • Auto launch - to auto load on start

  • No half vae - reduces distorted outputs

  • Xformers - boosts performance on Nvidia GPUs

  • Opt channels last - further performance gains on newer Nvidia cards

  • Mapv ram - reduces VRAM usage on <8GB GPUs

Save the file changes and close Notepad when finished. We're now ready to launch!

Installing the SDXL Refiner Extension

The SDXL Refiner model we downloaded earlier can be integrated into Automatic1111 via a handy browser extension. After Automatic1111 loads up for the first time, click on the Extensions tab and search for 'Refiner'. Click to install the extension, then visit the Install tab and click Check for Updates followed by Apply and Restart to finalize the integration.

This makes the SDXL Refiner available as a model option when you generate images later on. We can configure exactly how it gets used during each image generation, which allows us to fine tune quality vs. performance.

Generating Your First SDXL Image

We're finally ready to generate images! I'll briefly explain the key options to focus on:

  • Prompt - Enter text describing what you want to generate

  • Negative prompt - De-emphasize certain elements (e.g. NSFW)

  • Sampling method - Advanced option that impacts generation quality

  • Steps - More steps produce higher quality but take longer

  • Image dimensions - SDXL works best at 1024x1024 or above

  • Refiner model - The SDXL Refiner we integrated earlier

Try entering a prompt like 'a towering fantasy castle' and start generating images! Adjust the settings to optimize quality and performance to suit your specific needs. Have fun and let your imagination run wild!


Q: Why do I need to install Python 3.10.6 specifically?
A: Automatic1111 requires Python 3.10.6 to function properly. Other versions may not work correctly.

Q: What is Git and why do I need it?
A: Git is a version control system that lets you easily download, update, and manage software like Automatic1111 from GitHub.

Q: Where can I find more SDXL models?
A: You can find additional SDXL models at sites like Civetai.com, though some models may have NSFW content.

Q: How can I speed up SDXL image generation?
A: Using a graphics card with more VRAM can significantly speed up SDXL. See video for more tips on improving performance.

Q: What's the difference between batch size and batch count?
A: Batch size determines how many images generate simultaneously based on VRAM, while batch count is how many times to recreate the image.

Q: What should I set the CFG scale to?
A: A CFG scale between 5-7 balances listening to your prompts while still allowing some creativity from the AI.

Q: How does the refiner extension improve images?
A: The refiner uses an additional model on the final steps to enhance details.

Q: What resolution should I use for SDXL?
A: 1024x1024 is the standard for SDXL. Smaller sizes may produce low-quality images.

Q: Are there optimal settings for different use cases?
A: Yes, settings like sampling method and CFG scale affect the output. Adjust settings based on your specific needs.

Q: Can I revert to a previous version if I encounter issues?
A: Yes, Git allows you to easily revert back to older versions and commits if needed.