To Creativity, and Beyond: Empirical Studies for Learning Designs of Divergent and Critical Thinking with Text-to-Image Generative AI

Recent advances in generative AI have positioned text-to-image models as promising creativity support tools. These systems enable visually compelling outputs through intuitive interactions, offering new avenues for creative expression even for non-expert users. Particularly for children, such tools lower the barriers to creative expression by allowing them to visualize their ideas. However, most existing tools are designed for adults, relying on complex prompts and emphasizing output quality over the creative process, which may limit opportunities for children to engage in open-ended ideation, reflection, and critical evaluation-key aspects of creative learning. To address these challenges, we developed KidPrompt, a learner-centered text-to-image generative AI tool that incorporates three design components: (1) keyword-based input to simplify prompt construction, (2) choice induction to encourage comparison across multiple generations, and (3) sequential creation to support iterative idea development. A comparative study with 12 children suggests that these components can scaffold broader idea exploration, reflective comparison, and iterative refinement during creative interaction. Rather than emphasizing only polished outputs, KidPrompt encouraged children to explore multiple directions and revisit their ideas over time through structured yet flexible interaction. Based on these findings, we propose design implications for integrating generative AI into children’s creativity support in more learner-centered and process-oriented ways.