"I Recall the Past": Exploring How People Collaborate with Generative AI to Create Cultural Heritage Narratives

TL;DR

The paper addresses how individuals co-create personal cultural heritage narratives with generative AI, contrasting unofficial local voices with Authorized Heritage Discourse. It adopts a qualitative, workshop-based approach using Stable Diffusion to generate images of familiar and unfamiliar sites, extracting three narrative strategies and assessing AI strengths (illumination, amplification, reinterpretation) alongside limitations (detailed inaccuracies, cultural bias). The study contributes practical guidance on prompt engineering, model fine-tuning, and the use of objective references and curated datasets to improve cultural accuracy and inclusivity in AI-assisted storytelling. It highlights the need for AI to act as a supportive tool rather than a controller, and calls for broader dataset curation and ethical considerations to harness GenAI responsibly in cultural heritage contexts. Overall, the work demonstrates a viable path for inclusive, participatory heritage storytelling with GenAI while acknowledging and proposing mitigations for bias and knowledge gaps.

Abstract

Visitors to cultural heritage sites often encounter official information, while local people's unofficial stories remain invisible. To explore expression of local narratives, we conducted a workshop with 20 participants utilizing Generative AI (GenAI) to support visual narratives, asking them to use Stable Diffusion to create images of familiar cultural heritage sites, as well as images of unfamiliar ones for comparison. The results revealed three narrative strategies and highlighted GenAI's strengths in illuminating, amplifying, and reinterpreting personal narratives. However, GenAI showed limitations in meeting detailed requirements, portraying cultural features, and avoiding bias, which were particularly pronounced with unfamiliar sites due to participants' lack of local knowledge. To address these challenges, we recommend providing detailed explanations, prompt engineering, and fine-tuning AI models to reduce uncertainties, using objective references to mitigate inaccuracies from participants' inability to recognize errors or misconceptions, and curating datasets to train AI models capable of accurately portraying cultural features.

Figures (20)

• Figure 1: (a) is procedure diagram, (b) is the screenshot of the online workshop meeting.
• Figure 2: Three creation approaches used by participants, each corresponding to the images they ultimately generated. These three approaches encompass depicting realistic scenes, recreating memorable scenarios, and expressing exploratory creation.
• Figure 3: P20's generated images and prompts. She gradually incorporated many surreal elements into her prompts.
• Figure 4: P11's generated images and prompts evolved from her initial goal of depicting a building to gradually incorporating various narrative elements.
• Figure 5: P1's inpainting interface and the generated images. She tested a number of scenarios with different traffic volumes on the road and finally choose one from them.