How to run Stable Diffusion locally on your computer

Hardware requirements

The minimum requirement is a computer with a graphics card that has at least 6GB VRAM. It is possible to work on 4GB with some tweaks, but in my opinion, it’s not worth trying. The perfect scenario is to have a top-tier graphics card like Nvidia RTX4090. I recommend reading the article about the ways and costs of generating graphics before you start.

Installers

Some of popular User Interfaces (UIs) have their own installers. Even if that’s just a simple script that downloads and configures everything for you, it’s still an installer. It’s better than forcing users to execute commands manually in the terminal.

This is the list of popular UIs with installers:

• Forge
• ComfyUI
• Automatic1111
• InvokeAI
• Fooocus
• StableSwarmUI

You need to read the documentation of each installer to know how to use it. Some of them has significantly better performance than others. Some of them supports FLUX, some not yet. Each of them has different features, so you need to choose the one that suits your needs.

Cons:

• You may encounter problems with installation and running. Be prepared to spend some time on manual configuration and fixing errors.
• You may need to learn how to use Symbolic Links to share models between UIs.

Stability Matrix

Stability Matrix is a Multi-Platform Package Manager for Stable Diffusion. Supports Windows, Linux and MacOS. You can use it to download and run UIs locally without unnecessary configuration.

Pros:

• Easy to use.
• You don’t need to install requirements manually.
• You can have multiple UIs installed at the same time and share models between them.
• Has model browser that allows you to download models directly.

LynxHub

LynxHub is another Multi-Platform Package Manager for Stable Diffusion. Released for public in August 2024. Basically it does the same thing as Stability Matrix. Has a bit different features and UI. Here is a comment from the author about the differences between them. Personally, I didn’t have a chance to do large tests of LynxHub yet.

Pros:

• Easy to use.
• Nice looking User Interface.

Pinokio

Pinokio - is “a browser that lets you install, run, and programmatically control ANY application, automatically”. Pinokio is not focused on Stable Diffusion, but you can use it to run popular UIs. Community feedback is positive. However, due to the issues mentioned below, I have not tested it good enough yet.

Cons:

• Terrible App installer. You need to specify the path to install the app manually. And if that path already exists, you will need to delete it first.
• Has issues with App State Management. You can clearly see that when you change settings.
• Terrible UI/UX.

Docker

Docker is a platform that enables developers to automate the deployment of applications inside lightweight, portable containers. In easy words - you download an file that contains everything you need to run Stable Diffusion and you launch it. Anyway, it’s a bit more complicated than previous methods, because you will need to:

• Install and configure Docker.
• Select and download proper image from Docker Hub. Each image has different features and requirements. So you will need to spend some time on searching.
• Configure Container. For this part you will need to read the documentation of the image you downloaded. In most cases you will need to set up just a few environment variables.

Pros:

• You can switch between different Docker images.
• It’s secure. Things are isolated from your system. Even if you will run a Pickle Tensor model, it won’t harm your computer.

Cons:

• Sharing models between Docker images is harder. You will need to configure it.

There are tons of Docker images created by the community. You need to find the one that suits your needs. Check Docker Hub and GitHub. Personally, I recommend using AI-Dock images. They are well maintained and they give you a cool feature to use Provisioning Scripts to “perform certain actions when starting a container, such as creating directories and downloading files”.