Leveraging Vision-Language Models for Art Historical Analysis
Research Poster Arts & Humanities 2025 Graduate ExhibitionPresentation by Shuhua Yang
Exhibition Number 117
Abstract
Our research explores whether computer vision can detect artistic influence among 19th-century landscape painters. The key objectives are:(A) extract salient features of artistic style, (B) utilize quantitative and qualitative measures to compare artistic styles, and (C) use natural language to explain how models learn artistic style. To achieve this, we employ latent diffusion models for high-resolution image generation and style transfer, assessing their ability to capture the characteristics of realist paintings. By analyzing how diffusion models encode realism, we examine the intersection of computational and art historical perspectives. Additionally, we use LLMs for VAQ system designed for our painting dataset. By prompting the model to describe artistic style and generate formal art analysis, we explore AI’s ability to provide textual interpretations of artistic features. Our findings suggest that AI models can identify stylistic patterns and offer explanations, demonstrating their potential for studying artistic influence. However, they also highlight challenges in defining style computationally. This interdisciplinary approach connects computer vision and art history, providing new insights into how artistic styles evolve and influence one another.
Importance
This study bridges AI and art history, exploring how computer vision can detect artistic influence among 19th-century landscape painters. By analyzing paintings using advanced AI models such as diffusion models and LLMs, we discover patterns in artistic style that may not be visible to human observers. The use of machine learning allows for a new way to study how artists influenced each other, providing both visual and textual insights. The proposed study not only enhances our understanding of art but also pushes AI to interpret creative works in a more meaningful way. We intend to offer a novel perspective on how artistic styles evolve, making art analysis more accessible and systematic for researchers, educators, and enthusiasts alike.