Generative AI makes it increasingly easy to create illustrations for educational materials—but how can we systematically assess whether these images are actually suitable for learning?

In our new working paper, “Evaluation of AI generated illustrations for learning: A Practical Framework for BA Theses,” Sandra Schön, Benedikt Brünner and Martin Ebner present a practical framework for evaluating AI-generated educational illustrations.

The framework focuses on four core criteria: conceptual representation, visual correctness, legibility & labeling, and overall educational usability. It also provides guidance on target groups, learning objectives, prompting and standardized documentation to support reproducible and comparable research.

The framework was developed particularly with Bachelor’s thesis projects in mind and can serve as a practical starting point for systematically investigating the quality of AI-generated learning illustrations.

DOI: 10.3217/d4f2r-mft21 (TUGraz Repository)

Image generated with GPT Image 2