Direct Diffusion Score Preference Optimization via Stepwise Contrastive Policy-Pair Supervision

TL;DR

This paper introduces DDSPO, a novel method for aligning diffusion model outputs with user preferences by using dense, stepwise supervision derived from a pretrained reference model, reducing reliance on manual labels.

Contribution

DDSPO provides a new approach for preference optimization in diffusion models by automatically generating dense supervision signals from a pretrained model, improving alignment without manual annotations.

Findings

Outperforms existing preference-based methods in text-image alignment.

Requires less supervision than prior approaches.

Enhances visual quality of generated images.

Abstract

Diffusion models have achieved impressive results in generative tasks such as text-to-image synthesis, yet they often struggle to fully align outputs with nuanced user intent and maintain consistent aesthetic quality. Existing preference-based training methods like Diffusion Direct Preference Optimization help address these issues but rely on costly and potentially noisy human-labeled datasets. In this work, we introduce Direct Diffusion Score Preference Optimization (DDSPO), which directly derives per-timestep supervision from winning and losing policies when such policies are available. Unlike prior methods that operate solely on final samples, DDSPO provides dense, transition-level signals across the denoising trajectory. In practice, we avoid reliance on labeled data by automatically generating preference signals using a pretrained reference model: we contrast its outputs when conditioned on original prompts versus semantically degraded variants. This practical strategy enables effective score-space preference supervision without explicit reward modeling or manual annotations. Empirical results demonstrate that DDSPO improves text-image alignment and visual quality, outperforming or matching existing preference-based methods while requiring significantly less supervision. Our implementation is available at: https://dohyun-as.github.io/DDSPO