Kitap bölümü Açık Erişim

Comparison of Text-to-Image Generative AI Tools for Urban Portraits: A case study of Invisible Cities by using Stable Diffusion, DALL-E and MidJourney

2024    Kevseroğlu, Özlem; Kurban, Rifat

This book chapter compares the potential of text-to-image generative AI tools to realize urban portraits that are described through literary creation. Stable Diffusion XL Turbo, Midjourney V6 and DALL-E 3 (through the Microsoft Designer Image Creator and OpenAI, GPT-4) — in this case, written descriptions turned to visual— for five fictional cities from Italo Calvino's work, "Invisible Cities" are explored. A sample of 20 domain professionals scored the renderings obtained by the tools for their similarity to the core of Calvino's storytelling. The results indicate that the best tool is DALL-E 3, and within it, its version of GPT4, because it has excellent rendering skills for graphically very complex literary descriptions. The study discusses the potential for these AI tools to affect urban studies with the possibility of engaging new dialogue involving stakeholders and hopefully, fostering new approaches while enabling new areas.