On August 22, 2022, Stability AI announced the public release of Stable Diffusion in a post titled “Stable Diffusion Public Release.” The announcement stated that the release followed an earlier release to researchers and made the model broadly available, with the weights and code accessible through HuggingFace under the Creative ML OpenRAIL-M license, alongside an AI-based safety classifier.
The model is built on the latent diffusion technique introduced in the 2021 paper “High-Resolution Image Synthesis with Latent Diffusion Models” by Robin Rombach and colleagues. By performing image generation in a compressed mathematical space rather than directly on pixels, the method made high-quality text-to-image generation efficient enough to run on consumer-grade hardware.
Stable Diffusion was a landmark for open and accessible generative AI. Unlike closed commercial systems, its openly available weights let developers, researchers, and businesses run, study, and build on the model directly. This openness accelerated experimentation across the industry and made image generation a building block that any organization could adopt.