Only-Style: Stylistic Consistency in Image Generation without Content Leakage

2025-06-11

AI Summary

This work addresses semantic content leakage—a critical issue in reference-based style transfer for image generation. We propose a diffusion-based pure style transfer method that strictly preserves source content while transferring only stylistic attributes. Our contributions are threefold: (1) the first differentiable leakage localization mechanism coupled with a local adaptive style alignment paradigm, enabling patch-level style control and effective content-style disentanglement; (2) the first quantitative evaluation framework for measuring content leakage; and (3) a differentiable leakage score that drives dynamic parameter optimization during inference. Extensive experiments across multiple benchmarks demonstrate significant improvements over state-of-the-art methods: our approach reduces content leakage by up to 62% while maintaining high-fidelity style consistency, exhibits strong generalization across diverse styles and domains, and requires no additional training or fine-tuning.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Generation

Application Category

• Web Mining and Content Analysis: Content-based information diffusion
• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Generating images in a consistent reference visual style remains a challenging computer vision task. State-of-the-art methods aiming for style-consistent generation struggle to effectively separate semantic content from stylistic elements, leading to content leakage from the image provided as a reference to the targets. To address this challenge, we propose Only-Style: a method designed to mitigate content leakage in a semantically coherent manner while preserving stylistic consistency. Only-Style works by localizing content leakage during inference, allowing the adaptive tuning of a parameter that controls the style alignment process, specifically within the image patches containing the subject in the reference image. This adaptive process best balances stylistic consistency with leakage elimination. Moreover, the localization of content leakage can function as a standalone component, given a reference-target image pair, allowing the adaptive tuning of any method-specific parameter that provides control over the impact of the stylistic reference. In addition, we propose a novel evaluation framework to quantify the success of style-consistent generations in avoiding undesired content leakage. Our approach demonstrates a significant improvement over state-of-the-art methods through extensive evaluation across diverse instances, consistently achieving robust stylistic consistency without undesired content leakage.

Problem

• Prevent content leakage in style-consistent image generation
• Balance stylistic consistency and leakage elimination adaptively
• Quantify success in avoiding content leakage during generation

Innovation

• Localizes content leakage during inference
• Adaptively tunes style alignment parameter
• Proposes novel evaluation framework for leakage