Stability AI updates Stable Diffusion XL: version 0.9 available

Generative AI Stable Diffusion updates once again, but in its XL version. While more “geek” users can still use Stable Diffusion 2.0, version released by Stability AI towards the end of 2022, those who are subscribed to the beta of Stable Diffusion XL can download version 0.9 to try out the latest in image generation from text prompts .

The announcement of the update comes directly from the US company, which guarantees access to the model via ClipDrop starting today, June 23, 2023. The first open version should instead arrive in mid-July with the transition to the 1.0 iteration, allowing those with a modern consumer GPU (especially NVIDIA) to generate hyper-realistic images for film, television, music and educational videos, as well to templates for design and industrial use. Once again, SDXL places itself at the forefront of AI imaging applications.

Stability AI aggiorna Stable Diffusion XL

The update in question shows clear steps forward compared to the previous version of SDXL Beta. Update 0.9, as you can also see from the cover image, allows you to create much more realistic and detailed images, while also offering a range of features that extend beyond basic text suggestions. These include image-to-image prompting (inserting an image to get variations of that image), inpainting (reconstructing missing parts of an image), and outpainting (constructing a continuous extension of an existing image).

The key driver of this advancement for SDXL 0.9 is the significant increase in parameters on which the model is trained, compared to the beta version. In fact, there is talk of a base of 3.5 billion parameters from open source images, for a potential of 6.6 billion. SDXL 0.9 runs on two CLIP models, including one of the largest OpenCLIP models trained to date (OpenCLIP ViT-G/14), which boosts processing power and ensures the ability to create realistic images with greater depth and higher resolution. large size of 1024×1024.

SDXL 0.9 runs on systems with Windows 10, 11, or Linux, 16GB RAM, Nvidia GeForce RTX 20 graphics card (equivalent or better) with a minimum of 8GB VRAM, or even AMD 16GB VRAM for Linux users . The model is now available on Clipdrop, and Stability AI API and DreamStudio customers will be able to access it from Monday, June 26. It will initially remain available for research purposes only, prior to free dissemination on Github.

Source: Stability AI


Please enter your comment!
Please enter your name here