AI Startup Stability AI Unveils SDXL 1.0
AI startup Stability AI has recently launched its newest flagship text-to-image model, Stable Diffusion XL 1.0 (SDXL 1.0), boasting it as their most sophisticated model yet.
Enhanced Features and Accessibility
SDXL 1.0, with a staggering 3.5 billion parameters, is now available on GitHub, as well as through the Stability API and consumer apps like Clipdrop and DreamStudio. The latest version promises brighter and more accurate colors, superior contrast, shadows, and lighting compared to its predecessor.
Faster Image Generation with Versatility
According to Joe Penna, head of applied machine learning at Stability AI, SDXL 1.0 can create 1-megapixel images within seconds, supporting various aspect ratios. Unlike the previous Stable Diffusion XL 0.9 model, which demanded substantial processing power, SDXL 1.0 is more accessible, allowing it to run efficiently on simpler hardware.
Customizable and User-Friendly
Penna highlights that SDXL 1.0 is highly customizable and easily adaptable to different concepts and styles. With basic natural language processing queries, users can effortlessly craft complex designs.
Advancements in Text Generation
One of the noteworthy improvements in SDXL 1.0 lies in the domain of text generation. While most text-to-image models struggle with legible logos or calligraphy, SDXL 1.0 excels in advanced text generation and legibility.
Versatile Features for Detailed Variations
SDXL 1.0 offers a range of powerful features, including inpainting to restore missing image parts, outpainting to expand existing images, and image-to-image hints for adding multiple text hints after an image has been entered. Furthermore, the model effectively interprets complex multi-part instructions given in short prompts, eliminating the need for longer text prompts as required by previous Stable Diffusion models.
Addressing Concerns and Legal Challenges
The training set of SDXL 1.0 includes the works of artists who have expressed concerns over their artwork being used as training data for generative AI models by companies like Stability AI, notes NIXSolutions. Although Stability AI claims protection under the fair use doctrine, Getty Images sued the company for allegedly using its images illegally to train the generative neural network.