Zigzag Diffusion Sampling: Diffusion Models Can Self-Improve via Self-Reflection

TL;DR

This paper introduces Zigzag Diffusion Sampling, a self-reflection-based method that improves diffusion model outputs by leveraging the guidance gap between denoising and inversion, leading to higher quality and better prompt alignment.

Contribution

The paper proposes diffusion self-reflection and Z-Sampling, a novel, plug-and-play diffusion sampling method that enhances generative quality across various models and datasets.

Findings

Z-Sampling improves image quality and prompt alignment.

DreamShaper with Z-Sampling achieves up to 94% HPSv2 winning rate.

Z-Sampling enhances models like Diffusion-DPO with minimal costs.

Abstract

Diffusion models, the most popular generative paradigm so far, can inject conditional information into the generation path to guide the latent towards desired directions. However, existing text-to-image diffusion models often fail to maintain high image quality and high prompt-image alignment for those challenging prompts. To mitigate this issue and enhance existing pretrained diffusion models, we mainly made three contributions in this paper. First, we propose diffusion self-reflection that alternately performs denoising and inversion and demonstrate that such diffusion self-reflection can leverage the guidance gap between denoising and inversion to capture prompt-related semantic information with theoretical and empirical evidence. Second, motivated by theoretical analysis, we derive Zigzag Diffusion Sampling (Z-Sampling), a novel self-reflection-based diffusion sampling method that leverages the guidance gap between denosing and inversion to accumulate semantic information step by step along the sampling path, leading to improved sampling results. Moreover, as a plug-and-play method, Z-Sampling can be generally applied to various diffusion models (e.g., accelerated ones and Transformer-based ones) with very limited coding and computational costs. Third, our extensive experiments demonstrate that Z-Sampling can generally and significantly enhance generation quality across various benchmark datasets, diffusion models, and performance evaluation metrics. For example, DreamShaper with Z-Sampling can self-improve with the HPSv2 winning rate up to 94% over the original results. Moreover, Z-Sampling can further enhance existing diffusion models combined with other orthogonal methods, including Diffusion-DPO.