Stable Diffusion XL 1.0
SDXL 1.0 ships with open weights, a 3.5B-parameter base plus refiner and native 1024x1024 output, and runs on 8GB consumer GPUs.
UNet about 3x larger than earlier Stable Diffusion, a second text encoder, multi-aspect-ratio training and an optional refiner model for image-to-image polish. 6.6B parameters for the base+refiner ensemble. Released under the CreativeML OpenRAIL++-M licence on GitHub, Clipdrop, AWS and DreamStudio.
- Date
- Wednesday, 26 July 2023
- Lab
- Stability AI
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Parameters (base) | 3.5B 6.6B for the base+refiner ensemble | company |
| Minimum GPU memory | 8GB VRAM consumer-GPU claim | company |
Paper arXiv 2307.01952 (2023-07-04, Podell, Rombach et al.) preceded the 1.0 release; SDXL 0.9 was a research-only preview.
Sources
This record was checked against its sources on 6 October 2026. How we check