SDXL-Lightning is a significant project developed by ByteDance, utilizing a distillation strategy to quickly generate high-definition images. The model optimizes diffusion processes, achieving fast generation from text to high-resolution images. It adopts an innovative discriminator design and stable training techniques, ensuring high-quality generation and training stability. SDXL-Lightning has been open-sourced, supporting both LoRA and full UNet weight versions, suitable for various generation tasks. Through progressive distillation and adversarial distillation, it tackles issues of slow model speed and high computational costs.