Stability AI Debuts Photorealism-Focused Stable Diffusion XL Text-to-Image Generative AI Model
Synthetic media startup Stability AI introduced a new version of its Stable Diffusion text-to-image AI model aimed at enterprise customers. The new Stable Diffusion XL (SDXL) improves on several aspects of the standard Stable Diffusion model, with the company highlighting the photorealism of its images.
SDXL adds power and other enhancements to Stable Diffusion 2.1, the most recent iteration of Stability’s portfolio. The images generated from text prompts are more detailed and capable of looking more like photographs, with Stability pointing to particular advances in generating realistic faces. The AI is also better at filling in the details not mentioned in a prompt, allowing users to reduce the length of their descriptions without risking nonsensical and bizarre photo composition. SDXL can even impose recognizable and correctly spelled words within the images, something earlier versions of most text-to-image models have struggled with.
Stability has also brought editing and image generation tools to SDXL. The new image-to-image prompting feature will create variations of an image uploaded by a user as though it were one generated by the AI. Stability is also taking a page from OpenAI’s DALL-E text-to-image generator with the new inpainting and outpainting tools filling in incomplete images and extending the image beyond the existing dimensions, respectively. You can see how the inpainting tool works in the example below.
“We are excited to announce the latest iteration in our Stable Diffusion series of image solutions,” Stability AI CTO Tom Mason said. “SDXL brings a richness to image generation that is transformative across several industries, including graphic design and architecture, with results taking place in front of our eyes.”
Stability is running SDXL as a beta through its enterprise APIs and on its DreamStudio platform. The full release will be open source like the other models created by Stability. SDXL is already embedded in image-generating apps like NightCafe Studio.
The new model continues Stability AI’s recent streak of updates and improvements as it competes with new versions of Midjourney and other text-to-image generators. After raising $101 million last year, Stability has gone on to acquire the company behind AI image manipulation service Clipdrop and recently partnered with digital collectible platform Revel.xyz on an image-to-animation tool called Animai. Stability is also dealing with multiple potentially major legal allegations. A pending complaint in court from Getty Images and a class action lawsuit filed by a group of artists in California over claims of violated copyright and intellectual property laws haven’t slowed Stability’s development as of yet, however.
Follow @voicebotaiFollow @erichschwartz
Stability AI Unveils Image-to-Animation Generative AI Tool Animai
Stability AI Acquires Image Editing App Clipdrop Developer Init ML
New Midjourney V5 Generative AI Offers Photorealistic Images and Hands With 5 Fingers