SDXL Turbo, Stability AI’s new text-to-image mode, is now available. The model incorporates an innovative technique called Adversarial Diffusion Distillation (ADD), which allows for real-time text-to-image synthesis while maintaining high fidelity. The research paper detailing the ADD method is available for those interested in technical details. SDXL Turbo builds on the success of previous diffusion models and offers several advantages, including avoiding blurriness and artifacts commonly seen in other distillation methods. In performance tests, it outperformed other models with fewer computational requirements and improved inference speed. To experience the capabilities of SDXL Turbo, users can try it on Stability AI’s image editing platform, Clipdrop. While not currently intended for commercial use, interested parties can contact Stability AI for more information.
https://stability.ai/news/stability-ai-sdxl-turbo