Stable DiffusionStable Diffusion

In August 2022, Stable Diffusion was released with open weights, a decision that put a powerful text-to-image generator into the hands of anyone with a decent graphics card. Type a sentence, and out came an original image.

It was built on a technique called latent diffusion, introduced by researchers in Munich in the paper 'High-Resolution Image Synthesis with Latent Diffusion Models.' Diffusion models learn to create images by reversing a process of gradually adding noise, starting from pure static and refining it step by step into a picture.

The key efficiency trick was working in a compressed latent space rather than on full-resolution pixels. This slashed the computing power required, which is exactly what made it practical to run on consumer hardware rather than in a data center.

Because the weights were public, an enormous ecosystem sprang up almost overnight, with custom models, fine-tunes and editing tools built by a global community. It democratized image generation in a way closed systems had not.

It also intensified hard questions about the data these models learn from, the rights of artists whose work appeared in training sets, and the flood of synthetic imagery now reshaping the visual world.