An Image Generator Anyone Can Download and Run at Home

Stable Diffusion makes pictures from a text description, like the programs from OpenAI and Google, but it is free, its inner workings are public, and it runs on an ordinary gaming computer. Nobody controls what people make with it.

A London company called Stability AI released a program today called Stable Diffusion. It does what DALL·E 2 does — produces an image from a typed description — and the results are in the same league. The difference is what happens next. OpenAI's program lives on OpenAI's computers, behind a waiting list and a list of forbidden subjects. Stable Diffusion is a file you download. It runs on a home computer with a decent graphics card, the kind gamers buy. It has no waiting list, no usage rules, and no filter that can't be switched off.

The company says this is the point. Its founder, Emad Mostaque, has argued that a tool this powerful shouldn't be controlled by a few companies, and that putting it in everyone's hands is safer than locking it up. Programmers have already begun building on it — within hours, versions appeared that run in a web browser, on a phone, and inside photo-editing software.

Everything the closed programs prevent, this one allows. It will draw real people. It will draw violence. It will produce images in the specific style of living artists, whose work was among the billions of pictures it learned from and who were not asked. Those artists are, as of today, the loudest critics.

Whatever one thinks of the decision, it can't be undone. The file is on the internet. Making a convincing picture of anything has stopped being a service you're granted access to and become software you own.