Talk at 'From Hype to Reality: AI in the Study of Art and Culture'

Event: From Hype to Reality: AI in the Study of Art and Culture Date: 26 November 2025 Organizer: Digital Society Initiative · University of Zurich

---

Text-to-Image in the Wild: Ethics and Culture in Open-Source AI

Open-source text-to-image (TTI) pipelines currently dominate the landscape of AI-generated visual content. Personalization techniques such as LoRA enable users to fine-tune models for specific tasks, expanding both creative experimentation and harmful forms of visual synthesis.

Focusing on CivitAI, the most active platform for sharing and developing such models, this research presents a sociotechnical analysis of user-generated visual content and model development practices. Findings reveal a disproportionate rise in not-safe-for-work material and a significant number of models designed to mimic real individuals without consent.

While enabled by technical accessibility, these practices are also shaped by cultural patterns originating in earlier internet communities. Anime-focused imageboards and boorus developed collaborative tagging systems (“folksonomies”) that have since been incorporated into machine learning workflows via auto-captioning and dataset-tagging tools. These culturally contingent aesthetic categories are now embedded in the technical infrastructure of open-source AI.

Drawing on constructivist perspectives on technology, the talk situates open-source TTI within a broader genealogy of internet visual culture and examines how design choices, data infrastructures, and community norms contribute to exploitative visual media production.