Research Stay at Uppsala University

Institution: Uppsala University – Digital Humanities Department Date: 28 September – 5 October 2025

Text-to-image systems make it easy to generate images “in the style of” a given artist or genre. This notion of style is highly reductive, treating it as transferable statistical correlations, whereas art history understands style as a contested marker of context, authorship, and meaning.

Community-trained models tagged as style demonstrate how the term circulates primarily as a label for categorization and exchange rather than as a subject of art-historical debate. Fine-tuning experiments on hand-drawn data probe what models encode when trained on a unique drawing style, raising questions about what is captured as “style.”

While foundation models already distill style into tokens, community-trained personalizations such as LoRA create highly specific, distributable modules that commodify individual (mostly digital) artists’ styles. Framed as democratization, this tokenization risks weakening style as personal signature and unsettles notions of authorship, consent, and ownership.

Workshop: Synthesizing Style: Authorship and Consent in Text-to-Image AI Personalizations