This project aimed to integrate a large language model (LLM) and text-to-image generator into an existing game, "Run and Find," to create an educational tool by creating a Pipeline. The study examined the diversity of the Pipelines generated game stories and evaluated the Pipeline’s textual output’s educational and engagement value. Teacher interviews indicated an interest in using games for education but raised concerns about the learning structure and student engagement. To address these concerns, the game was modified to include educational checkpoints, requiring players to answer questions to progress, reinforcing learning outcomes. The study employed a double-diamond methodology for iterative development and testing. DALL-E 3 and Pipedream as a development environment, were chosen for their compatibility with the existing game’s visual style and implementation efficiency, and ChatGPT 4 was chosen as LLM for the Pipeline. Prompt engineering was utilized to craft system messages that would tailor the output to align with the game structure. The surveys conducted with teachers and teacher students gauged the game’s educational value. The result of the survey was positive. The Pipelinegarnered interest from teachers, who believed it would be engaging and educational for students, especially with its adaptability in content generation and editing options. While the chosen methodology suited the project, further theoretical investigations and testing phases could enhance results. The stories generated with the same prompt had a similar diversity as stories generated with different prompts, indicating good diversity.