Invited Talk – AISthesis Network, Aarhus University

Created on February 12, 2026

2026

Event: AISthesis – Network for Research on the Aesthetics of AI Imagery Date: 12 February 2026 Host: Aarhus University

Cultural Inscriptions and Taxonomies in Text-to-Image Models

This study examines how open-source text-to-image systems inherit and reproduce cultural taxonomies shaped by Web2 infrastructures and aesthetics. Focusing on anime-focused imageboards (“boorus”), the research analyzes how collaborative tagging systems (“folksonomies”) became embedded in machine learning workflows through tools such as DeepDanbooru and WD-14 Tagger.

Although originally pragmatic dataset solutions, these vocabularies now persist and evolve far beyond anime contexts, shaping generative AI aesthetics at scale.

By tracing this path dependency, the study highlights how contingent historical data practices become entrenched within generative systems, influencing semantics, style, and bias in contemporary machine-generated imagery.