Stable Diffusion XL 1.0: the open-source quality jump
In one sentence Stability ships SDXL 1.0 (3.5B base + 6.6B refiner), native 1024×1024 output, shorter prompts. Open source under commercial license, weights on HuggingFace.
Exactly one year after Stable Diffusion 1, Stability AI ships SDXL 1.0. Images now come out at 1024×1024 pixels by default (was 512×512), with much sharper detail, hands with the right number of fingers more often, and text in images that's almost readable.
Key shift: prompts work much better even short. On SD 1.5 you had to write "8k masterpiece highly detailed cinematic lighting trending on artstation" to get anything decent. On SDXL "a photo of a black cat on a red couch" is enough.
Weights are public on Hugging Face with a commercial license, so anyone can download, modify, and sell results. Over the next 4 weeks Civitai hosts thousands of fine-tunes and LoRAs, and ComfyUI explodes as a node-based alternative to Automatic1111.
Companies
Stability AI
Tools
Stable Diffusion XL 1.0, SDXL
Tags
Sources