Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Jiang, Zhou, Wen, Yandong, Liu, Zhen
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2602.18799
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918349621952512
author Jiang, Zhou
Wen, Yandong
Liu, Zhen
author_facet Jiang, Zhou
Wen, Yandong
Liu, Zhen
contents Aligning large-scale text-to-image diffusion models with nuanced human preferences remains challenging. While direct preference optimization (DPO) is simple and effective, large-scale finetuning often shows a generalization gap. We take inspiration from test-time guidance and cast preference alignment as classifier-free guidance (CFG): a finetuned preference model acts as an external control signal during sampling. Building on this view, we propose a simple method that improves alignment without retraining the base model. To further enhance generalization, we decouple preference learning into two modules trained on positive and negative data, respectively, and form a \emph{contrastive guidance} vector at inference by subtracting their predictions (positive minus negative), scaled by a user-chosen strength and added to the base prediction at each step. This yields a sharper and controllable alignment signal. We evaluate on Stable Diffusion 1.5 and Stable Diffusion XL with Pick-a-Pic v2 and HPDv3, showing consistent quantitative and qualitative gains.
format Preprint
id arxiv_https___arxiv_org_abs_2602_18799
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Rethinking Preference Alignment for Diffusion Models with Classifier-Free Guidance
Jiang, Zhou
Wen, Yandong
Liu, Zhen
Computer Vision and Pattern Recognition
Aligning large-scale text-to-image diffusion models with nuanced human preferences remains challenging. While direct preference optimization (DPO) is simple and effective, large-scale finetuning often shows a generalization gap. We take inspiration from test-time guidance and cast preference alignment as classifier-free guidance (CFG): a finetuned preference model acts as an external control signal during sampling. Building on this view, we propose a simple method that improves alignment without retraining the base model. To further enhance generalization, we decouple preference learning into two modules trained on positive and negative data, respectively, and form a \emph{contrastive guidance} vector at inference by subtracting their predictions (positive minus negative), scaled by a user-chosen strength and added to the base prediction at each step. This yields a sharper and controllable alignment signal. We evaluate on Stable Diffusion 1.5 and Stable Diffusion XL with Pick-a-Pic v2 and HPDv3, showing consistent quantitative and qualitative gains.
title Rethinking Preference Alignment for Diffusion Models with Classifier-Free Guidance
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.18799