Stable Diffusion XL

Stable Diffusion XL, usually shortened to SDXL, is Stability AI's open-weight latent diffusion model for generating 1024x1024 images from text. It uses a two-stage pipeline: a base model produces the initial image latents and an optional refiner model sharpens fine detail, with prompts interpreted through a pair of text encoders, OpenCLIP-ViT/G and CLIP-ViT/L. The weights live in the Stability-AI generative-models repository alongside SDXL-Turbo, a distilled variant that generates in very few steps using adversarial diffusion distillation, and the Stable Video Diffusion models. Released under the CreativeML Open RAIL++-M license, SDXL runs locally on GPUs with around 8 GB of VRAM and remains one of the most broadly supported checkpoints in the open image ecosystem, with fine-tunes, LoRAs, and ControlNets available across tools like ComfyUI and the AUTOMATIC1111 web UI. Typical users range from hobbyists generating art locally to studios building image features on open weights.