Stable Diffusion SDXL 0.9: A Game Changer in AI Artistry
Written on
Introduction to SDXL 0.9
Following the successful debut of the Stable Diffusion XL (SDXL) beta in April 2023, Stability AI has introduced the highly anticipated SDXL 0.9. This latest iteration brings significant enhancements in image detail and composition when compared to its earlier versions.
Here are a few sample images produced with SDXL 0.9:
Image Quality and Comparisons
The left image was prompted with: "An aesthetic manicured hand holding a take-out coffee, pastel chilly dawn beach Instagram film photography," while the negative prompt included terms like "3D render, smooth, plastic, blurry, grainy, low-resolution, anime." The right image prompt was: "A wolf in Yosemite National Park, chilly nature documentary film photography," with similar negative prompts.
More examples can be found in their official announcement on Discord.
From the showcased examples, it is evident that the quality now rivals that of MidJourney. A comparison shared by Twitter user @amli_art highlights this:
Prompt: "A painting by the artist of the dream world, in the style of hybrid creature compositions, intricate psychedelic landscapes, hyper-realistic bird studies, colorful moebius, weirdcore, pink and cyan, cybermysticpunk."
The left image is from MidJourney, and the right one is from SDXL 0.9. Both outputs are visually striking. Which one do you prefer?
Advancements in Functionality
The SDXL series goes beyond basic text prompting, offering advanced features such as image-to-image prompting, inpainting, and outpainting.
Technical Enhancements
The key factor driving the advancements in composition for SDXL 0.9 is its considerable increase in parameter count over the beta version. It features one of the highest parameter counts among open-source image models, with a 3.5 billion-parameter base model and a 6.6 billion-parameter ensemble pipeline.
SDXL 0.9 operates on two CLIP models, including one of the largest OpenCLIP models trained to date, enhancing its processing capabilities and enabling the creation of realistic images with greater depth and a resolution of 1024x1024.
Accessing SDXL 0.9
SDXL 0.9 is currently available on Clipdrop. Additionally, developers and users can access the model via the Stability AI API and DreamStudio starting June 26, 2023. Please note that the Clipdrop platform was temporarily down as of June 24, 2023, but is expected to be operational again by June 26, 2023.
Alternatively, users can run the model locally, provided their systems meet the following specifications:
- Windows 10, 11, or Linux operating systems
- 16GB of RAM
- Nvidia GeForce RTX 20 graphics card with at least 8GB of VRAM
- Linux users may utilize a compatible AMD card with 16GB VRAM
Conclusion and Future Prospects
In summary, the rapid advancement of Stable Diffusion's open-source models is commendable, now nearing the quality of premium text-to-image models like MidJourney. Currently, the model is available for research purposes only, with interested researchers encouraged to apply for access. A full open release of SDXL 1.0 is anticipated soon, with expectations for even greater enhancements in the consumer release!
For developers keen on leveraging AI image generators for web applications, I highly recommend utilizing the Stable Diffusion API. It offers improved quality at a more competitive price compared to its counterparts.
Please consider supporting my work on Medium, where you can gain unlimited access by becoming a member through my referral link. Have a wonderful day!
Stay connected with us on LinkedIn for the latest updates in AI and let's shape the future of this exciting field together!
