WaDi: Weight Direction-aware Distillation for One-step Image Synthesis

TL;DR

This paper introduces WaDi, a novel one-step image synthesis method that uses weight direction analysis and low-rank rotations to improve distillation efficiency and performance in diffusion models.

Contribution

We propose the LoRaD adapter and integrate it into VSD to create WaDi, achieving state-of-the-art results with fewer trainable parameters and broad downstream task generalization.

Findings

WaDi achieves top FID scores on COCO datasets.

LoRaD models directional weight changes efficiently.

Distilled models generalize well to various tasks.

Abstract

Despite the impressive performance of diffusion models such as Stable Diffusion (SD) in image generation, their slow inference limits practical deployment. Recent works accelerate inference by distilling multi-step diffusion into one-step generators. To better understand the distillation mechanism, we analyze U-Net/DiT weight changes between one-step students and their multi-step teacher counterparts. Our analysis reveals that changes in weight direction significantly exceed those in weight norm, highlighting it as the key factor during distillation. Motivated by this insight, we propose the Low-rank Rotation of weight Direction (LoRaD), a parameter-efficient adapter tailored to one-step diffusion distillation. LoRaD is designed to model these structured directional changes using learnable low-rank rotation matrices. We further integrate LoRaD into Variational Score Distillation (VSD), resulting in Weight Direction-aware Distillation (WaDi)-a novel one-step distillation framework. WaDi achieves state-of-the-art FID scores on COCO 2014 and COCO 2017 while using only approximately 10% of the trainable parameters of the U-Net/DiT. Furthermore, the distilled one-step model demonstrates strong versatility and scalability, generalizing well to various downstream tasks such as controllable generation, relation inversion, and high-resolution synthesis.