run your own image AI with StableDiffusion automatic1111

You find AI photo editing inside phones, inside photo editing software, local and online. In addition there’s a way to create images from an idea in words – aka ‘generative AI’. You can make a logo or illustrate a newsletter page or something really fancy. You can also feed this AI your own photo, alter it dramatically, increase the resolution, in-paint an item (eg add banana) or outpaint (eg extend the image beyond its cropped edge.

Stable Diffusion is quietly famous – it may be at the backend of commercial services. A key point that will make sense later is that Stable Diffusion is the shorthand name for the bit, aka ‘the model’ that contains the AI imagery. It’s the result of a lot of training on the world’s photos to mightily p*s off professionals. There are other models which you’ll hear about as they are launched by Google or Meta. There are components (lora; VAE) which are accessories you use to influence the result eg ‘xx is good at human faces; yy is good at cartoonery’ – but here we just want to get started ahead of developing our skills.

want to DIY?

Because I have daily uses for image-AI, I’ll explain below how I set about installing it ‘for free’ on a pretty good PC made for gaming. AI is mostly not free – and it’s used as an incentive to get you to upgrade your phone or laptop or sell you a premium service such as Lightroom or Photoshop or any of a list of apps and online services. These ways are probably better and will save your time.

some options for doing image-AI

• for some AI features and the least effort, get to know your phone camera or phone photo gallery editor to find what AI can do.
• go play at https://stablediffusionweb.com online – presently a freemium service.
• install STABLE DIFFUSION to use locally – as described below. You’ll gain highly flexible, offline and very technical tools and do plenty of image-AI.
• You can save a lot of time by instead installing EASY DIFFUSION on the best computer available. If Stable Diffusion is a SLR camera, Easy Diffusion is a snapshot camera. This (‘ED’) installs like any single app and it downloads as it goes taking an hour or so. The easy requirement is to have many GB available for the ‘models’ that it will download. The other need (when I wrote this in 2025) was to have good NVIDIA graphics internals. A lot of the stuff that is downloaded for ED can be transferred to use in Stable Diffusion.

how to install Stable Diffusion

Installing Stable Diffusion on Docker on Linux or Windows WSL went wrong for me many times over. This is just partly due to competence – for example, when NVIDIA changes the name of its driver software, this subtle change causes downloads to go wrong and waste the time of thousands of people. There was no one click install. On my nth attempt I got Google’s Gemini AI to guide me through to a successful install. Me and my friend ‘Gemini’ started by talking about the equipment we’d need – should I use what I had or something new? ‘Gemini’ helped me decide to use my unused Lenovo Legion 7, a gamers laptop which had a fancy NVIDIA RTX graphics card. In short, I asked Google’s Gemini to give me a step by step guide. After each problem I pasted a chunk of on-screen messages for Gemini to say whether all was fine, what was wrong and what new thing needed to be done at that point. The help provided was fantastic to the point when Gemini told me that Stable Diffusion was in fact already running and I needed to view it using a URL.
The long story is that this took a few hours … so to be useful to you I have collected a lot of the errors and tips in the block below. Google Gemini summarised the process, the errors and the fixes and then created the HTML summary that you see now.

the step by step, snag by snag guide to a Stable Diffusion/Automatic1111 interface

let’s compare ways to install Stable Diffusion

The approach with docker compose is recommended for ease of management on Windows with WSL2 and NVIDIA GPUs and Apple Silicon Macs with MPS.

this works: Docker Compose with abdBarho repository 

• This method uses Docker Desktop (running on Apple Silicon Macs OR Windows with WSL2) to create an isolated container environment. The AbdBarho/stable-diffusion-webui-docker repository provides docker-compose.yml files that define how to build and run the Automatic1111 Web UI within this container. Your local files (models, outputs) are “mounted” into the container so you can access them with your usual file manager.
• Advantages:
  • isolation: It packages all the necessary Python versions, libraries (like torch, gradio), and dependencies inside the container separate from your main Windows or WSL2 Ubuntu environment. This largely eliminates “dependency hell”.
  • consistency: The application runs in the same environment every time, regardless of changes to your host system.
  • GPU acceleration: Docker Desktop’s WSL2 integration and the NVIDIA Container Toolkit efficiently pass through your RTX GPU to the container for high performance.
  • cleanliness: No messy Python installations directly on your system. You just manage Docker images and containers.
  • easy management: docker compose up and docker compose down provide simple commands to start and stop the entire application stack.

this is tricky: direct/native install (e.g. via Conda)

• This involves installing Python (often via Conda), Git, and all of Automatic1111’s dependencies directly onto your Windows machine or directly into your WSL2 Ubuntu distribution.
• Disadvantages:
  • dependency conflicts (python 3.10 issue): This was the source of a ImportError: cannot import name ‘TypeIs’ from ‘typing_extensions’. Different Python applications often require specific versions of libraries, leading to conflicts and a very frustrating experience.
  • environment pollution: your system becomes cluttered with Python packages and binaries.
  • manual management: requires more manual setup and troubleshooting of environment variables, CUDA paths, and Python virtual environments.
  • less portable: Moving the setup to another machine is much harder than with Docker.

use a pre-built Docker image? (e.g. kestr3l/stable-diffusion-webui)

• This is also a Docker-based approach, but typically uses a single docker run command to pull and run a specific pre-built image.
• Advantages: benefits from Docker’s dependency isolation and cleanliness. Can be very quick to get started if the image is well-maintained and matches your needs.
• Disadvantages:
  • less customization: If you need to make changes to the underlying environment (like the typing_extensions fix we did), you might have to rebuild the image manually or look for a different pre-built one.
  • maintenance: The specific image (kestr3l/stable-diffusion-webui:1.2.2 in your example) might not be as actively updated or community-supported as a general Docker Compose setup like AbdBarho’s.
  • complex docker run commands: A single docker run command can become very long and unwieldy with many volume mounts and environment variables, making it harder to read and manage than a docker-compose.yml file.

yes, this works on MPS Macs 

1. Intel Macs (Older):
   • Native/Conda: Possible, but very slow as they lack NVIDIA CUDA GPUs. They would fall back to CPU or potentially OpenCL (for AMD GPUs), which is significantly less performant.
   • Docker: Docker Desktop runs, but GPU acceleration via NVIDIA CUDA is not possible as Intel Macs don’t have NVIDIA GPUs. It would still run, but slowly on the CPU inside the container.
2. Apple Silicon Macs (M1, M2, M3, etc.):
   • Native/Conda: Yes, Automatic1111 has a well-optimized backend that leverages Apple’s Metal Performance Shaders (MPS). This can provide very good comparable performance. A direct install using Conda for an MPS-optimized environment is a common way to run it on Apple Silicon.

• Docker: Docker Desktop runs excellently on Apple Silicon. You can run Automatic1111 in Docker on these Macs. However, to get GPU acceleration, the Docker image itself needs to be built or configured to utilize MPS. The AbdBarho repository specifically offers mps profiles in its docker-compose.yml that are designed for Apple Silicon GPUs, making it a viable Docker-based option for Macs.