Generative AI-Driven Construction of the "Urban Cultural Image Map": A Multimodal Narrative Communication Model for Creative City Governance

The strategic governance of creative cities is frequently hindered by fragmented and siloed cultural data, which significantly impedes the synthesis of actionable narratives for effective decision making. This study addresses this critical gap by proposing and rigorously evaluating a generative AI-driven model for constructing an "Urban Cultural Image Map." This innovative model leverages multimodal large language models (MLLMs) to seamlessly integrate heterogeneous data types-including extensive textual archives, complex visual imagery, and detailed geospatial information-into a unified, highly structured knowledge graph. Through a comprehensive three-phase methodology encompassing data structuring, narrative generation, and practical governance application, the research develops and tests a dynamic platform that systematically organizes cultural assets into thematic narrative layers. Findings demonstrate that MLLMs can effectively perform cross-modal alignment, generate coherent thematic narratives, and support spatial pattern identification for complex governance tasks. However, the model exhibits certain limitations in generating deep contextual analysis, handling contested cultural meanings, and providing direct prescriptive policy insights without human oversight. The study concludes that while generative AI serves as a powerful augmentative tool for data synthesis and narrative communication, its successful integration into urban governance fundamentally requires a hybrid intelligence approach. Ultimately, this research contributes a novel framework at the intersection of urban studies, media communication, and artificial intelligence, offering practical pathways for developing more intelligent, narrative-sensitive tools for creative city governance.