The Year in AI Papers
The year/Independent research

Paper 2508.18966

USO: Unified Style and Subject-Driven Generation via Disentangled and Reward Learning

Shaojin Wu, Mengqi Huang, Yufeng Cheng, Wenxu Wu, Jiahe Tian, Yiming Luo, Fei Ding, Qian He

Published
Aug 2025
Research lab
Independent
Citations
29
GitHub
1.2K stars

01 In brief

Summary

The paper introduces USO, a unified framework for style-driven and subject-driven image generation, arguing these tasks are complementary and can be jointly optimized through cross-task co-disentanglement.

The authors construct a large-scale triplet dataset (content, style, stylized images) using a subject-for-style curation framework with stylization and de-stylization experts.

USO employs a two-stage training: style alignment training using SigLIP and a hierarchical projector, followed by content-style disentanglement training with separate encoders.

A style reward learning (SRL) paradigm further enhances performance.

The authors release USO-Bench, a benchmark for evaluating both style similarity and subject fidelity.

Experiments show USO achieves state-of-the-art results on subject-driven, style-driven, and joint tasks, with highest CLIP-T, DINO, and CSD scores on USO-Bench and DreamBench.

Ablations confirm the effectiveness of each component.

02 From the paper

Abstract

Existing literature typically treats style-driven and subject-driven generation as two disjoint tasks: the former prioritizes stylistic similarity, whereas the latter insists on subject consistency, resulting in an apparent antagonism. We argue that both objectives can be unified under a single framework because they ultimately concern the disentanglement and re-composition of content and style, a long-standing theme in style-driven research. To this end, we present USO, a Unified Style-Subject Optimized customization model. First, we construct a large-scale triplet dataset consisting of content images, style images, and their corresponding stylized content images. Second, we introduce a disentangled learning scheme that simultaneously aligns style features and disentangles content from style through two complementary objectives, style-alignment training and content-style disentanglement training. Third, we incorporate a style reward-learning paradigm denoted as SRL to further enhance the model's performance. Finally, we release USO-Bench, the first benchmark that jointly evaluates style similarity and subject fidelity across multiple metrics. Extensive experiments demonstrate that USO achieves state-of-the-art performance among open-source models along both dimensions of subject consistency and style similarity. Code and model: https://github.com/bytedance/USO