Stable Diffusion web UI
Want to make or fix pictures with AI without a monthly fee? This browser page lets you create images from a sentence, extend or repair photos, and enlarge them on your own computer.
How you use it: Install Python and git, run the install script for your computer, then use the page it opens in your browser.
In plain English
The README calls it a web interface for Stable Diffusion. You type a prompt and get an image, or you start from an image you already have. The interface runs through Gradio, which is a Python way to put controls in the browser.
The feature list is concrete. It includes text-to-image and image-to-image, inpainting, outpainting, upscaling, face restoration with GFPGAN or CodeFormer, a negative prompt for things you do not want to see, and the ability to save the settings inside the picture. It also says a 4 GB video card is supported, and it mentions reports of 2 GB cards working.
The install note is easy to miss: there is a one-click script, and you still must install Python and git. GitHub's last push for this repository was March 2, 2026. It was not marked archived. The license GitHub lists is AGPL-3.0.
Practical ways to use it
Example 1
Make a picture from a sentence
The README's first feature is text-to-image. You describe the picture, pick a sampler if you want, and generate.
Example 2
Change an existing picture
Image-to-image, inpainting, and outpainting are all named. Inpainting fills a region you mask. Outpainting extends past the edges.
Example 3
Repair a face or enlarge a picture
The extras tab lists GFPGAN and CodeFormer for faces, and RealESRGAN and other upscalers for resolution.
Example 4
Reuse the exact recipe
The README says generation parameters are saved in PNG or JPEG files, and you can drag the image back to restore them.
Who it is for
Some tech comfort. The screen is a page on your computer, which feels approachable. Getting there still means installing Python and git, even for the one-click script.
Difficulty: Takes some setup. Windows, Linux, and Apple silicon each have install sections. A video card helps. The one-click path is not a store install.
Getting started
- 1
Install Python and git first
The README says the one-click script still requires Python and git. Install those before you download the project.
- 2
Follow the install that matches your machine
The README has a Windows NVIDIA section, automatic Windows and Linux sections, and an Apple silicon section. Use the one for your computer.
- 3
Start the page and keep it local
After the script runs, it serves a page on your machine. Use that local address. Do not expose it to the internet.
- 4
Notice how recent the project is
The last GitHub push we recorded was March 2, 2026. If a model you care about is newer than that, check whether this UI still supports it.
What you need first
- Accounts
- None for the app itself. Model files may come from other sites, and those sites sometimes ask for an account. This README does not require a paid API key.
- Hardware
- The README says 4 GB video card support, with reports of 2 GB working. The README centers on GPU installs.
- Cost
- You run it on your hardware. The code license is AGPL-3.0. Model files have their own licenses. There is no price in the README because this repository does not sell a hosted plan.
Cautions
- You must install Python and git. The phrase one-click in the README does not remove those tools.
- The repository was last pushed on March 2, 2026. Treat newer model advice on the web as something to verify.
- AGPL-3.0 is a strong copyleft license. There is also a switch, --allow-code, that lets the UI run arbitrary Python. Leave that off.
Questions people ask
What does Stable Diffusion web UI do?
It gives you a local browser page to generate and edit images with Stable Diffusion, including inpainting, upscaling, and face restoration.
Who is it for?
People who can install Python and git and who want a visual page rather than a node graph or a Python script.
What do I need first?
Python, git, and a video card. The README documents 4 GB card support and separate instructions for Windows, Linux, and Apple silicon.
What should I watch out for?
The one-click script still needs Python and git. The project was last updated on March 2, 2026. Leave the allow-code switch off.