Stable Diffusion is a family of text-to-image models: each takes a written prompt and generates a picture. The first version came from the CompVis research group in collaboration with Stability AI and Runway; Stability AI Ltd now develops and licenses the family.
It is a latent diffusion model. A diffusion model starts from random noise and removes the noise step by step until an image that fits the prompt remains; a latent model does this on a compressed version of the image, which needs less computing. Stable Diffusion v1 has an 860-million-parameter UNet and a CLIP text encoder, was trained on 512x512 images from the LAION-5B data set, and runs on a GPU with at least 10 GB of memory. Because the weights are public, a business can run the model on its own servers and fine-tune it on its own photos. Later versions are Stable Diffusion XL and Stable Diffusion 3.5 in Large, Turbo and Medium sizes.
Stability AI offers the models for self-hosting, free under its Community License for organizations with less than US$1 million in yearly revenue, and through a paid API.