Step-level Denoising-time Diffusion Alignment with Multiple Objectives

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Qi, Wang, Dawei, Zou, Shaofeng
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910133067448320
author Zhang, Qi
Wang, Dawei
Zou, Shaofeng
author_facet Zhang, Qi
Wang, Dawei
Zou, Shaofeng
contents Reinforcement learning (RL) has emerged as a powerful tool for aligning diffusion models with human preferences, typically by optimizing a single reward function under a KL regularization constraint. In practice, however, human preferences are inherently pluralistic, and aligned models must balance multiple downstream objectives, such as aesthetic quality and text-image consistency. Existing multi-objective approaches either rely on costly multi-objective RL fine-tuning or on fusing separately aligned models at denoising time, but they generally require access to reward values (or their gradients) and/or introduce approximation error in the resulting denoising objectives. In this paper, we revisit the problem of RL fine-tuning for diffusion models and address the intractability of identifying the optimal policy by introducing a step-level RL formulation. Building on this, we further propose Multi-objective Step-level Denoising-time Diffusion Alignment (MSDDA), a retraining-free framework for aligning diffusion models with multiple objectives, obtaining the optimal reverse denoising distribution in closed form, with mean and variance expressed directly in terms of single-objective base models. We prove that this denoising-time objective is exactly equivalent to the step-level RL fine-tuning, introducing no approximation error. Moreover, we provide numerical results, which indicate our method outperforms existing denoising-time approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2604_14379
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Step-level Denoising-time Diffusion Alignment with Multiple Objectives
Zhang, Qi
Wang, Dawei
Zou, Shaofeng
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Reinforcement learning (RL) has emerged as a powerful tool for aligning diffusion models with human preferences, typically by optimizing a single reward function under a KL regularization constraint. In practice, however, human preferences are inherently pluralistic, and aligned models must balance multiple downstream objectives, such as aesthetic quality and text-image consistency. Existing multi-objective approaches either rely on costly multi-objective RL fine-tuning or on fusing separately aligned models at denoising time, but they generally require access to reward values (or their gradients) and/or introduce approximation error in the resulting denoising objectives. In this paper, we revisit the problem of RL fine-tuning for diffusion models and address the intractability of identifying the optimal policy by introducing a step-level RL formulation. Building on this, we further propose Multi-objective Step-level Denoising-time Diffusion Alignment (MSDDA), a retraining-free framework for aligning diffusion models with multiple objectives, obtaining the optimal reverse denoising distribution in closed form, with mean and variance expressed directly in terms of single-objective base models. We prove that this denoising-time objective is exactly equivalent to the step-level RL fine-tuning, introducing no approximation error. Moreover, we provide numerical results, which indicate our method outperforms existing denoising-time approaches.
title Step-level Denoising-time Diffusion Alignment with Multiple Objectives
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.14379