Investigating How AI Can Support the Creation of Storyboards for Autism Education

This Master’s dissertation investigates how generative AI can support the creation of clear, consistent, and autism-appropriate storyboard images for education. A modular image generation pipeline was developed using Stable Diffusion XL (SDXL), integrating IP-Adapter for identity preservation, ControlNet for spatial guidance, and Compel for extended prompt encoding. The system was designed to produce low-complexity, prompt-faithful visuals tailored to the cognitive needs of autistic learners. A novel evaluation framework combined BLIP-Large for semantic alignment, CLIP for visual consistency, and computer-vision-based metrics for artifact and complexity detection. Quantitative testing demonstrated substantial improvements in visual simplicity (97% classifier accuracy), prompt alignment (87% semantic accuracy), and consistency (+11.7% gain with IP-Adapter). Qualitative feedback from special education professionals confirmed that the AI-generated storyboards met autism-informed design principles of clarity, predictability, and emotional readability. This research provides one of the first integrated frameworks uniting diffusion-based image generation with educational accessibility, showing that AI can enhance inclusivity and scalability in autism-focused learning materials