Reconstructing Lost Urban Memory with Generative AI: An Oral-History Prompt-Engineering Framework

Urban transformation processes not only destroy the physical fabric of historic neighbourhoods but also erode the collective memory and everyday life practices embedded within them. This study addresses this spatial memory loss by employing generative AI to visually reconstruct the lost atmospheres of Cumhuriyet Neighbourhood within the outer walls of Kayseri Castle, Turkey. Drawing on oral history interviews conducted as part of a TÜBİTAK-funded project, narrative data from 5 former residents were transformed into structured seven-category prompts and fed into Google Gemini- Nano Banana 2 pro, OpenAI GPT-Image-1, and XAI Grok. A 30-member residents panel evaluated the outputs. Tool G (Google Gemini) achieved the best overall ranking (M = 1.77). The findings confirm that human curatorial oversight and iterative community feedback remain essential for culturally authentic AI-generated heritage visualisations.