Conference Agenda

This paper explores an AI-assisted approach to iconographic analysis of medieval manuscript illuminations. Adapting a workflow originally developed for early modern woodcuts - combining Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) - we apply it to automatically generate Iconclass metadata for the 14th-century Wenzelsbibel. Our study makes two main contributions: assessing the method’s suitability for medieval visual culture and evaluating its integration into a TEI-based digital edition. We critically reflect on challenges such as limited domain-specific training data and stylistic variation across periods. Central to our approach is an "expert-in-the-loop" design ensuring scholarly oversight. This work aligns with the conference’s theme by addressing both the algorithmic foundations for visual data analysis and the epistemological implications of machine-generated metadata in the humanities.