Experiments in the Relationship between Art History and Text-to-Image Models

This chapter investigates how text-to-image models such as DALL-E, Midjourney, and Stable Diffusion interpret prompts derived from art historical analysis. Drawing on formal and iconographic methods, it applies a black-box strategy to evaluate how these models respond to canonical artworks across diverse media and periods. Through comparative analysis of generated images, the study highlights the models' capacity for stylistic emulation, as well as persistent gaps in representing symbolism, historical context, and themes of female power and identity. The findings reveal cultural biases embedded in training data and underscore the models' limitations in grasping complex iconographic language and symbolic meaning. Text-to-image models, the chapter argues, are not only tools for replication or illustration, but should also be critically examined and explored for their potential as frameworks for art history research.