From Words to Wonder: Designing and Evaluating an AI-Empowered Creative Storytelling System for Elementary Children

Research Background and Issues

• What problems or challenges did the authors identify?
  While there are many digital creative storytelling tools for children, the design of tools based on Generative AI remains underexplored, particularly in enhancing story quality (e.g., creativity and richness). Additionally, traditional visual thinking tools (e.g., generating visual images or story scripts) overly rely on children's visual thinking abilities, which makes it difficult to express abstract narratives or complex story structures. This is especially limiting for older children, hindering the development of their storytelling skills.

• Why is this issue important?
  Creative storytelling is a critical learning activity that enhances elementary school children's narrative abilities, language literacy, socio-emotional development, and creativity. However, most existing storytelling tools struggle to balance textual and visual prompts and lack research on their effectiveness in improving story quality. This presents a significant research opportunity in the field of education.

• Research Motivation and Related Work
  The research is motivated by the potential of Generative AI for personalized and dynamic creation. Existing tools, such as StoryDrawer and AI Story, focus on AI-based visual storytelling but are limited to single-mode prompts. They do not fully explore how multimodal (text + image) generative AI can support children's creative expression at complex narrative levels.

Solution

• What methods or solutions did the authors propose?
  The authors proposed an interactive system called StoryPrompt, which integrates text and visual generative AI to help elementary school children enhance their language and creative abilities through co-creating stories and comics. It supports children in balancing story structure planning with incremental detail adjustments by providing inspiration through keyword and image generation.

• What are the innovative aspects of this solution?

  • Integration of multimodal prompts: Combines text (generated keywords) and visuals (AI-generated images), enabling children to benefit from both linguistic prompts and visual elements to spark their imagination.
  • Support for planning and emergent creativity: Allows children to preset story structures while expanding story details through randomly or contextually generated AI prompts.
  • Dynamic keyword generation: Real-time generation of keywords with varying degrees of relevance based on children's input to stimulate story creation.
  • Interactivity and practicality: Provides adaptive designs for students of different grades, allowing younger students to use the system with parental or teacher support.

• What are the implementation steps and key technologies used?

  1. Children initially select main characters, emotional trajectories, and story settings to plan the preliminary story structure.
  2. The system uses ChatGPT to generate keywords related to the children's input (with varying levels of relevance), enabling them to associate the keywords with the story context.
  3. After children create textual content, the AI generates comic-style images using the Stable Diffusion model.
  4. The story is organized into a six-page comic format, which children can download or review.
     Key technologies include ChatGPT, the Stable Diffusion model for image generation, and system construction and interaction design based on Unity.

Research Outcomes

• What specific results were achieved?

  1. Experimental results showed that compared to traditional paper-based storyboarding methods, StoryPrompt significantly improved the quality of children's stories, particularly in creativity and richness.
  2. During the creation process, children were better able to purposefully plan their overall stories and explore alternative narrative content through AI-generated keywords and images.
  3. Students from different grades showed positive acceptance and preference for the system, with AI-generated images being the most popular feature.
  4. Teachers provided feedback suggesting that the system has the potential to stimulate learning motivation and can be flexibly integrated into classroom teaching.

• What advantages does it have compared to existing solutions?

  • Combines textual and visual prompts, expanding the capabilities of previous AI storytelling systems that focused on a single modality.
  • Enhances the complexity and narrative freedom of generated content, making it more suitable for the needs of children across different grades.
  • Balances planning and emergent details, making it easier for children to develop higher-level narrative skills.

• What were the experimental or evaluation results?

  • Story Quality: Stories generated with the AI system significantly outperformed those created using traditional methods in terms of word count, creativity, and richness. The average word count per story increased from 180 words (traditional method) to 249 words, and creativity scores improved from 2.87 to 3.45.
  • Time and Order Effects: In the study design, the AI system consistently outperformed traditional methods. Even after controlling for variables (e.g., usage order), the AI's effects remained significantly different.
  • Children's Preferences: All children expressed willingness to continue using StoryPrompt, with the integration of AI-generated images and keywords being the most favored features.
  • Teacher Feedback: Teachers found the system highly engaging for students and suggested improving the precision and consistency of image outputs.

• Limitations and Future Directions

  1. Experimental Design Limitations: Insufficient time was allocated to evaluate older students' ability to create longer narratives, and no longitudinal studies were conducted to verify the sustainability of learning outcomes.
  2. Neglect of Individual Differences: The study did not thoroughly explore how differences in age or background influenced children's use of the AI tool in their creative processes.
  3. Technical Limitations: The output of generative AI (e.g., image accuracy) did not fully meet all children's expectations, limiting the creative expression of some participants.

Conclusion

This study systematically investigated how to design and evaluate a generative AI-supported creative storytelling system for elementary school students, offering new insights for the design and practice of educational technology for children. By integrating the advantages of text and visual generative AI and balancing creative planning with emergent details, StoryPrompt demonstrated the potential of generative AI in children's creative education. However, future research could deepen the study of individual differences, optimize the precision and consistency of generated content, and conduct long-term follow-up experiments to more comprehensively understand the impact of generative AI on children's creative development.