对抗扩散蒸馏

Nov, 2023

Adversarial Diffusion Distillation

Axel Sauer, Dominik Lorenz, Andreas Blattmann, Robin Rombach

TL;DRAdversarial Diffusion Distillation (ADD) is a new training approach that efficiently samples large-scale image diffusion models in 1-4 steps, outperforming existing few-step methods and reaching state-of-the-art performance in only four steps, enabling real-time image synthesis.

Abstract

We introduce adversarial diffusion distillation (ADD), a novel training approach that efficiently samples large-scale foundational image diffusion models in just 1-4 steps while maintaining high image quality. We