Generative AI and Science Storytelling: A Roadmap for Multimedia Content Creation for Science Education

Abstract

The learning of abstract scientific concepts presents a significant comprehension challenge, requiring a pedagogical shift toward transforming complex data into accessible visual narratives. While storytelling helps students conceptualize invisible phenomena (e.g., microscopic of physical processes), the manual production of these multimodal resources remains a severe operational barrier for teachers, Generative Artificial Intelligence (GenAI) offers a scalable solution, however, its integration is currently hindered by the models lack of logical reasoning capabilities, their tendency to generate visual hallucinations (factual inaccuracies), and the varying AI literacy levels among educators. To address these barriers, this research asks: How can a storytelling-based design roadmap be developed to assist educators to systematically guide GenAI in producing scientifically valid and narratively coherent science education content? The main goal is the development of a Storytelling-Based Design and Production Roadmap, a structured framework meant to guide educators through the human-AI co-creation process. Methodologically this dissertation is based on a Design-Based Research (DBR) approach, following the benchmarking of current GenAI tools, the Roadmap was prototyped into a dual-tier architecture based on the DigCompEdu framework. The artifact underwent empirical validation with a triangulated sample of pre-service science educators and AI/multimedia specialists, utilizing qualitative friction logs and quantitative evaluations grounded in the Aristotelian Triad and Alberto Cairo's Visualization Wheel. This work contributes a validated operational framework that guides educators through GenAI generation via template-driven scaffolding (Tier 1) and a conversation verification loop (Tier 2). The results demonstrate that systemic prompt engineering is effective at suppressing AI hallucinations and decreasing visual noise, the study ultimately provides concrete pedagogical design principles to bridge the gap between educator intent and probabilistic machine outputs and shows that structured human oversight is critical in producing multimedia science narratives that are both engaging and scientifically rigorous.