Launch of Stable Diffusion XL Version 1.0: Enhanced Artistic Capabilities Unveiled

DALL-E, Midjourney, and other similar tools helped kick off the generative AI wave last year, but Stable Diffusion stood out for its ability to run locally on your PC without the use of external servers. Stable Diffusion XL, an upgraded model, has now left beta and into “stable” territory with the arrival of version 1.0.

Stability AI, the company behind Stable Diffusion, said, “SDXL 1.0 is the flagship image model from Stability AI and the best open model for image generation. We’ve tested it against various other models, and the results are conclusive - people prefer images generated by SDXL 1.0 over other open models. This research results from weeks of preference data captured from generations of experimental models on our Discord and from external testing.”

Stability AI says that SDXL 1.0 is showing promising results in the wild today, and that in various tests, users preferred images generated by SDXL 1.0 over images made by other models. SDXL 1.0 is seemingly able to surpass its predecessor in rendering notoriously challenging concepts, including hands, text, and spatially arranged compositions.

One of SDXL 1.0’s outstanding features is its architecture. It has a 3.5B parameter base model and a 6.6B parameter refiner, making it one of the most parameter-rich models in the wild right now. It has a two-stage architecture that involves a base model and a refinement model specialized for denoising — and this can apparently provide amazing results without having to compromise speed or requiring excessive computational resources. And that versatility extends to custom data fine-tuning, making it easier than ever for users to tailor the model to their specific needs.

If you want to check out Stable Diffusion XL 1.0, there are several ways you can get started with it. It’s live on Clipdrop, and it’s available as an image generation model on DreamStudio. It’s also live on GitHub if you would like to peruse through its source code. It’s definitely an interesting development for AI — Stable Diffusion was already one of the best models out there for image generation, and this should mean good news for those that want to use AI to generate images.

Source: Stability AI