Stability AI has recently unveiled its most advanced AI image generator, known as Stable Diffusion XL 1. 0. The company has made this artificial intelligence (AI) tool available in open source on GitHub as well as on its consumer apps, ClipDrop and DreamStudio. Boasting an impressive 3. 5 billion parameters, Stable Diffusion XL (SDXL) 1. 0 has the capability to generate high-quality, one-megapixel images in various aspect ratios. Parameters serve as the model's learned inputs from the training data, essentially defining the program's skill level. According to Stability AI, the refining process of SDXL ensures that it produces images with vibrant and accurate colors, improved contrast, lighting, and shadows compared to its predecessor. Additionally, the company claims that the new model is capable of understanding simpler language prompts, reducing the need for complex instructions. Stability AI also offers a beta feature for fine-tuning, enabling the generation of personalized images of specific individuals and products using a small set of images. However, this feature is currently undergoing limited testing and is scheduled for release in the following weeks. Furthermore, Stability AI highlights improvements in text generation with the latest Stable Diffusion model. While text-to-image generators often struggle to produce legible text, Stability AI insists that SDXL excels in advanced text generation.

Leaked images of an alleged DALL-E 3 have also demonstrated enhanced text generation capabilities. SDXL supports both inpainting and outpainting techniques, wherein inpainting involves altering specific sections of an image and outpainting involves extending the image beyond its original borders in a coherent manner, akin to Adobe's Generative Fill tool. The open-source nature of Stable Diffusion raises concerns about its potential misuse by bad actors for generating harmful content, such as nonconsensual deepfakes. Tech Crunch points out the availability of numerous tutorials on using Stable Diffusion for creating deepfakes and explicit content. It is worth noting that SDXL's training data includes works from artists who object to their creations being utilized in this manner and claim that Stability AI lacks proper authorization to do so. The legal implications surrounding fair use doctrine and potential lawsuits against the company remain to be seen. In response, Joe Penna, Stability AI's head of applied machine learning, emphasizes the company's commitment to improving safety features and complying with artists' requests for their removal from training datasets. SDXL will also be prominently featured on Amazon Webrock, a cloud platform hosting various generative AI models. Emad Mostaque, CEO of Stability AI, describes the release of 1. 0 on Amazon Bedrock as a testament to the company's dedication to collaborating with AWS in delivering state-of-the-art open access models to the AI community and providing optimal solutions for developers and clients.