Stability AI Releases Powerful New Image Generator Stable Diffusion XL
Stability AI has announced its new AI image generator Stable Diffusion XL 1.0 which the company describes as its “most advanced” model to date.
The artificial intelligence (AI) tool is available in open source on GitHub and on its consumer apps ClipDrop and DreamStudio.
Stable Diffusion XL (SDXL) 1.0 contains 3.5 billion parameters and can generate one-megapixel images in multiple aspect ratios. Parameters are what the model learns from the training data and later used as inputs; it essentially defines the skill of the program.
Stability AI says that the refining process of SDXL means it will generate “more vibrant and accurate colors, with better contrast, lighting, and shadows than its predecessor.” It also says that prompts need be less complicated and it will understand simpler language.
Stability API offers a new fine-tuning beta feature that is capable of taking a small set of images to generate images of specific people and products. However, that feature is currently in limited testing and will be released in the coming weeks.
The company says the latest Stable Diffusion has improved in the area of text generation. Text-to-image generators often struggle to produce eligible text but Stability AI insists SDXL is capable of “advanced” text generation. Yesterday, leaked images from a rumored DALL-E 3 also showed better text generation.
SDXL also supports inpainting and outpainting. Inpainting is changing part of an image while outpainting means extending outside of an original image in a coherent manner, similar to Adobe’s Generative Fill tool.
Controversies
The open-source nature of Stable Diffusion means it can be used by bad actors to generate harmful content, like nonconsensual deepfakes. Tech Crunch notes there are countless tutorials on how to use Stable Diffusion to create deepfakes and generate porn.
SDXL was trained on data produced by artists who resent their work being used in this way, claiming they never gave companies like Stability AI permission to do so. The company says it is shielded by the fair use doctrine but that is yet to be tested in a court and the company is facing lawsuits.
“We are constantly improving the safety functionality of Stable Diffusion and are serious about continuing to iterate on these measures,” Joe Penna, Stability AI’s head of applied machine learning, tells Tech Crunch. “Moreover, we are committed to respecting artists’ requests to be removed from training data sets.”
SDXL will also be featured on Amazon Webrock, a cloud platform that hosts generative AI models.
“The latest SDXL model represents the next step in Stability AI’s innovation heritage and ability to bring the most cutting-edge open access models to market for the AI community,” says Emad Mostaque, Chief Executive Officer of Stability AI.
“Unveiling 1.0 on Amazon Bedrock demonstrates our strong commitment to work alongside AWS to provide the best solutions for developers and our clients.”
