SDXL in 4 steps with Latent Consistency LoRAs — 10x faster inference

AI Impact Summary

Stable Diffusion XL (SDXL) image generation speed has been significantly improved by introducing Latent Consistency Model (LCM) LoRAs. This allows for inference in approximately 1 second on an M1 Mac with SDXL, a 10x speedup compared to the standard 7-second generation time on a 3090. This change enables real-time event applications and dramatically reduces iteration times for artists and researchers, unlocking new use cases and workflows.