SDXL Turbo

The real-time text-to-image model that generates high-fidelity outputs in a single step.

SDXL Turbo is a fine-tuned, distilled version of the foundational SDXL 1.0 model, engineered for unprecedented speed. It leverages a novel training method: Adversarial Diffusion Distillation (ADD), enabling image synthesis in a single network evaluation step. This single-step capability drastically reduces the required step count from 50 to just one, achieving state-of-the-art performance. On an A100 GPU, SDXL Turbo generates a 512x512 image in approximately 207ms, delivering a major improvement in inference speed while maintaining high image quality and fidelity.

https://stability.ai/news/stability-ai-sdxl-turbo