Stability AI, the creator of the popular prompt-based AI image generator Stable Diffusion, has announced its latest version, SDXL 0.9, which promises to deliver significantly more photorealistic images and improved hand rendering. The announcement, although briefly featured in a now-deleted blog post, has been reported by Bloomberg, generating excitement among AI enthusiasts and creators.
According to Stability AI, SDXL 0.9 represents a remarkable leap in creative possibilities for generative AI imagery. Sample images showcased in the blog post demonstrate the notable advancements achieved when comparing the same prompts in the previous beta version, Stable Diffusion XL, and the new SDXL 0.9. The images generated using the latest model exhibit finer details and more convincing hand representations, a significant improvement from earlier iterations. Hands, which were often a telltale sign of AI-generated art, have become increasingly realistic, particularly with the emergence of Midjourney v5, a competing platform running on Discord, launched earlier this year.
Stability AI just released SDXL 0.9. The quality of it is insane and it's free/open source.
Comparison between SDXL 0.9 (left) and Midjourney 5.1 (right)
prompt: portrait of a young woman, blue eyes, cinematic
It's available on the Clipdrop, Stability AI API and DreamStudio pic.twitter.com/iTyHJAMmtu— Kris Kashtanova (@icreatelife) June 22, 2023
The team at Stability AI attributes the enhanced results to a significant increase in the model’s parameter count, which encompasses all the weights and biases within the neural network used for training. The larger parameter count enables SDXL 0.9 to produce more refined and lifelike creations. While the model can run on a standard home computer, utilizing the software locally will require a minimum of 16GB of RAM and a GeForce RTX 20 (or higher) graphics card with 8GB of VRAM. It is compatible with Windows 11 / 10 and Linux operating systems.
In the deleted blog post, Stability AI mentioned plans to integrate SDXL 0.9 into their Clipdrop web tool and incorporate it into their DreamStudio app. Additionally, the startup revealed that an open-source version of SDXL 1.0 is scheduled for release in mid-July, further expanding accessibility and opportunities for creative applications.
The advancements showcased by SDXL 0.9 hold great promise for a wide range of industries. Stability AI believes the improved capabilities will find practical use in film, television, music, and instructional video production, as well as making significant contributions to design and industrial applications. The real-world potential of SDXL places it at the forefront of AI imagery, driving innovation and creative expression.
As the official release date approaches, the AI community eagerly awaits the arrival of SDXL 0.9 and looks forward to the broader accessibility and transformative possibilities it brings to the realm of generative AI imagery.