Stable Diffusion 1.5

A latent diffusion model with ~859M parameters that generates images from text. It works by creating images in a compressed space (32×40) and then expanding them to full size (512×512) using a decoder. It's trained on ~2B curated images from LAION-Aesthetics and uses OpenAI's CLIP text encoder to understand prompts. Released in August 2022, it was the standard for text-to-image generation for about a year.