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

TL;DR

This paper investigates how AI art models unintentionally reproduce and blend artistic styles without explicit prompts, introducing a new evaluation protocol called Art Arena to measure this phenomenon.

Contribution

The paper formulates evaluation principles for style leakage in AI art generation and introduces Art Arena, a protocol to quantify style reappearance in generated outputs.

Findings

Style leakage varies with representational strength and interaction dynamics.

Models show asymmetric blending of styles in generated images.

Art Arena effectively measures unintended style reappearance.

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.