An image model you can own
Stable Diffusion, released in 2022, is an image generation AI whose weights were published openly. Anyone can download it free and run it on their own machine, and that freedom is what broadened image generation so quickly. The Diffusion in the name refers to the diffusion model underneath.
What being open enables
- Free and local: generate without limit, and without sending images anywhere
- Customization: additional training produces models for your own style or characters
- Ecosystem: improved models and tooling published by people worldwide
- Embedding: companies can build it into their own services
It feels less like using an official service and more like acquiring tools and setting up your own workshop, which is why it appeals to committed creators, researchers, and companies.
Compared with Midjourney
Midjourney and the image generation in chat assistants win on convenience and hit rate; Stable Diffusion wins on freedom and volume. Running it locally needs a reasonably capable machine — a GPU in particular — and some setup, so it is less a first step than the next step for someone who has become serious about image generation.
Context worth having
Being open has also meant continuous debate about misuse and infringement. Litigation over training data and measures against inappropriate generation were argued around Stable Diffusion before anywhere else. It stands as the product that showed both sides of open source models.
Stability AI went through a period of reported turmoil, but the models and tooling that spread through the community remain active, with successors and derivatives continuing to appear.