The Silent Brush: Evaluating Artistic Style Leakage in AI Art Generation

2026-05-17

AI Summary

This work addresses the risk that AI-generated models may implicitly reproduce copyrighted or stylistically distinctive artworks without explicit prompting, raising concerns about attribution and intellectual property. The paper introduces the concept of the “Silent Brush,” formally defining and quantifying this non-explicit style leakage for the first time. It proposes Art Arena, a generalizable evaluation framework that integrates diffusion models—including Stable Diffusion v1.5, SDXL, and SANA-1.5—to systematically assess implicit style leakage through style encoding strength measurement, cross-prompt reproduction detection, and interactive modeling of artistic works. Experiments reveal significant disparities in how different artworks are represented within models, leading to uneven leakage patterns. Art Arena effectively identifies and quantifies these phenomena, establishing a new benchmark for copyright compliance and model transparency in generative AI.

Abstract

Generative text-to-image models are typically trained on large-scale web-scraped datasets that include diverse visual content such as copyrighted and stylistically distinctive artworks, raising concerns about ownership, attribution, and the unintended reuse of protected visual expressions. A key issue is that models can learn stylistic patterns from this data and reproduce them in generated outputs without any explicit reference in the prompt. We refer to this phenomenon as The Silent Brush, where such learned styles reappear even when they are not requested. Existing evaluation methods mainly focus on near-duplicate retrieval or membership inference and do not account for this form of unintended stylistic resurfacing across prompts. To address these gaps, we first formulate guiding principles for evaluation of The Silent Brush. We then introduce Art Arena, an evaluation protocol that measures how strongly artworks are encoded, how they interact, and how frequently their stylistic traits reappear in generated outputs without explicit mention in prompts. We evaluate Art Arena on widely used text-to-image diffusion models, including Stable Diffusion v1.5, Stable Diffusion XL (SDXL), and SANA-1.5, and design it to generalize across text-to-image generative systems. Our results show that The Silent Brush arises from differences in representational strength and interaction dynamics between artworks, leading to asymmetric blending in model generations. Code and evaluation resources are available at: https://anonymous.4open.science/r/ArtArena-EBE4.

Problem

• style leakage
• AI art generation
• copyright
• text-to-image models
• artistic style

Innovation

• Silent Brush
• Art Arena
• style leakage
• text-to-image generation
• diffusion models