In this guide

A guide to Stable Diffusion

Stable Diffusion is a collection of open-source models by Stability AI. They are used to generate images, most commonly as text to image models: you give it a text prompt, and it returns an image. But they can also be used for inpainting and outpainting, image-to-image (img2img) and a lot more.

In this guide we’ll show you the basics, then dive deeper into using these models to get the best out of them.

If you haven’t used Stable Diffusion before, give it a go now. Try asking for a painting of a cute cat:

Predicted image

There are many different versions of Stable Diffusion. Let’s break it down:

  1. Stable Diffusion XL (SDXL) is currently the most popular version of Stable Diffusion. It was released in July 2023 and creates fantastic, realistic images around the 1024x1024 resolution, though you can use any aspect ratio.
  2. Stable Diffusion 1.5 (SD1.5) is an older version that was open-sourced in August 2022, and its images are best at 512x512. Despite its age, it remains popular because of its speed, low memory usage, and an abundance of community fine-tuned models which use SD1.5 as a base.
  3. Stable Diffusion 2.1 (SD2.1) was released in October 2022, and was “sort of like the awkward second album”. Good, just… different. This version introduced improvements like negative prompts, OpenCLIP for a text encoder, larger image outputs, but the migration to OpenClip caused significant changes to the image output and composition, when compared to prior versions of Stable Diffusion. To many, this felt like a “breaking change”. Most notably, this migration caused many artist’s names to be removed from the text encoder, which to this day drives a subset of users to use 1.5 or SDXL over version 2.
  4. SDXL Turbo is a version of SDXL that was released in November 2023. It is a non-commercial model that is very fast and can make good images in a single step.
  5. SD Turbo is another fast and non-commercial version that was released in November 2023.

How to use Stable Diffusion →

Run AI with an API.

Run and fine-tune open-source models. Deploy custom models at scale. All with one line of code.

Try Replicate