Stability AI, the startup funding a range of generative AI experiments, has released a new version of Stable Diffusion, the text-to-image AI system that was among the first to rival OpenAI’s DALL-E 2. This version, named Stable Diffusion XL (SDXL), offers improvements upon the original in several aspects, such as graphic design and architecture. Tom Mason, Stability AI’s CTO, mentioned that the new system has brought a “richness” to image generation that the previous model lacked.
SDXL is also better at generating hands, an anatomy that previous text-to-image systems struggled to recreate due to its complexity. The hands generated by SDXL are more realistic than the nightmare fuel generated by the predecessor, Stable Diffusion 2.1.
Moreover, SDXL is supposedly better at generating text, although testing results could still use some improvements. It also features enhanced image composition and face generation, apart from straightforward text-to-image prompting input. Additionally, SDXL has outpainting and inpainting functions, while its predecessor only handles prompting.
Once it exits the beta stage, SDXL will be open-sourced. It is already accessible through Stability’s API and its generative art tool DreamStudio, in early access.
Stability AI has been met with legal challenges due to how its tools have been commercialized and built, with Getty Images suing the company for reportedly using its images without permission for the original Stable Diffusion. The open source release of Stable Diffusion has also caused controversy since it has few restrictions, allowing some communities to run pornographic celebrity deepfakes and graphic depictions of violence. Authorities have called for regulation to address these models.
Stability AI has responded by pledging to honor artists’ requests to have their art removed from the training data of the next-generation Stable Diffusion models aka “Stable Diffusion 3.0.” To date, over 78 million works have been removed. The company is seeking to monetize its AI efforts, which encompass a variety of things such as art and animation, biomed and generative audio. CEO Emad Mostaque has expressed an interest on potential IPO, though it is yet to generate substantial revenue.
Stability AI is a startup providing a range of generative AI experiments. Founded by Tom Mason, their CTO, the company uses text-to-images AI systems, such as their new version, the Stable Diffusion XL (SDXL). SDXL is more competent when it comes to generating graphics, hands and text, as well as image-to-image prompting, inpainting and outpainting. The open source release of Stable Diffusion has become controversial due to its lack of restrictions, and Getty Images has taken the company to court for using their images without permission. As for monetization efforts, Emad Mostaque has hinted at IPO chances.
