Skip to content

Stability AI Releases SDXL, a Text-to-Image AI System That Outshines OpenAI’s DALL-E 2

Stability AI, a startup that's been shaking up the world of generative AI, has announced the release of Stable Diffusion XL (SDXL), the latest and greatest version of its groundbreaking text-to-image AI system. SDXL has been developed to rival OpenAI's DALL-E 2, and it promises to be a game-changer for industries like graphic design and architecture. With improved image composition, better face generation, and the ability to create "descriptive imagery" without lengthy prompts, SDXL represents a huge leap forward in the field of generative AI.

While SDXL can recreate complex objects such as hands better than its predecessor, it still has limitations in generating text, a challenging task for generative AI models. The new system is reportedly on par with or even better than the latest release of its competitor, Midjourney's model.

Stability AI's foray into generative AI art has drawn legal scrutiny, with allegations that the company infringed on millions of artists' rights by developing its tools with copyrighted images from web scraping. Stock image supplier Getty Images has also taken the company to court for using its images without permission to create the original Stable Diffusion model. Additionally, the open-source release of Stable Diffusion has been controversial, with concerns over the model's usage restrictions leading to its use in generating pornographic celebrity deepfakes and graphic violence depictions. These controversies have raised questions about the need for regulation to address models like Stable Diffusion that don't moderate content sufficiently.

Despite these challenges, Stability AI is keen to monetize its AI efforts in various fields, including art, animation, biomedicine, and generative audio. While CEO Emad Mostaque has hinted at an IPO, reports suggest that the company is burning through cash and slow to generate revenue, despite raising over $100 million in venture capital funding in October 2022.