Multistep Distillation of Diffusion Models via Moment Matching

TL;DR

This paper introduces a novel multistep distillation technique for diffusion models that significantly accelerates sampling while maintaining or improving quality, achieving state-of-the-art results on Imagenet and high-resolution image generation.

Contribution

It extends one-step diffusion distillation to multi-step models using moment matching, enabling faster sampling with superior performance.

Findings

Distilled models with up to 8 steps outperform their one-step and original multi-step counterparts.

Achieved state-of-the-art results on Imagenet dataset.

Demonstrated fast high-resolution image generation without autoencoders.

Abstract

We present a new method for making diffusion models faster to sample. The method distills many-step diffusion models into few-step models by matching conditional expectations of the clean data given noisy data along the sampling trajectory. Our approach extends recently proposed one-step methods to the multi-step case, and provides a new perspective by interpreting these approaches in terms of moment matching. By using up to 8 sampling steps, we obtain distilled models that outperform not only their one-step versions but also their original many-step teacher models, obtaining new state-of-the-art results on the Imagenet dataset. We also show promising results on a large text-to-image model where we achieve fast generation of high resolution images directly in image space, without needing autoencoders or upsamplers.