Stable Diffusion 1.5
Stability AI · August 2022
Why It Matters
Broke open the AI image generation field by being the first high-quality model anyone could download and run for free, creating an entire ecosystem of tools, extensions, and custom models.
Description
The open-source model that democratized AI image generation. Uses a technique called latent diffusion — generating images by working in a compressed mathematical space rather than pixel-by-pixel, which made it fast enough to run on consumer GPUs. Trained on billions of image-text pairs from the LAION-5B dataset, it spawned a massive community of artists, developers, and hobbyists.
Notable Milestones
- ▸Enabled local AI image generation on consumer hardware for the first time
- ▸Spawned thousands of community-created custom models and extensions
- ▸Foundation for ControlNet, LoRA fine-tuning, and other community innovations
Key Innovations
Family Tree
Successors (1)
Related Research (2)
Showed that gradually adding noise to data and then learning to reverse the process could generate images rivaling GANs, with more stable training and…
Trained a model to understand both images and text by learning which image-text pairs go together from 400 million internet examples. This created a s…