As digital technologies increasingly reshape how cultural memory is shared and preserved, this paper proposes an exploratory design framework that fuses Large Language Models (LLMs) with multimodal interactive systems to revitalise intangible cultural heritage through real-time, context-aware storytelling. Using Guangzhou’s Spring Flower Market as a case study, we integrate an LLM-driven semantic engine with TouchDesigner and Kinect sensing to translate textual or spoken cultural prompts into visually coherent, emotionally resonant generative scenes. The approach addresses persistent limitations of gesture-only installations—shallow narrative depth and fragmented visual style. By allowing the LLM to interpret developer intent and audience input, our system dynamically assembles node graphs that maintain cultural symbolism while adapting to user interaction, thereby lowering technical barriers for artists and enhancing visitor engagement. We outline the system architecture, present a prototype deployment, and discuss ethical considerations surrounding AI-mediated heritage narration.