EXPLORING THE ACCURACY AND USABILITY OF AI-GENERATED IMAGES FOR ABSTRACT PHYSICS CONCEPTS IN EDUCATION
Yoseph Satria Praka, Malisa Huzaifa, Candi Rahayu Tri Prameswari, Nurul Wahyuni, Ananda Maria

Digital Multimedia Engineering, Jakarta State Polytechnic, Depok, Indonesia

Abstract

Abstract physics concepts are often difficult for students to understand due to their
invisible and complex nature, requiring effective visualization. Recent advances in artificial
intelligence (AI) enable the rapid generation of educational images- however, their scientific
accuracy and usability remain uncertain. This study aims to evaluate the accuracy and usability
of AI-generated images for selected abstract physics concepts using Google Gemini with a
CLEAR-based prompting approach. This study employed a mixed methods descriptive evaluative
design, combining quantitative Likert-scale ratings and qualitative expert feedback. The
evaluation involved expert judgment from a physics teacher with 11 years of teaching experience.
The results show that the accuracy of AI-generated images is generally high (mean scores: 3.75
for kinetic theory of gases and 3.5 for atomic structure), although limitations were found in
representing complex conceptual relationships and maintaining scientific consistency. In contrast,
usability scores were higher (4.25 and 4.0), indicating that the images are visually appealing and
helpful for learning. The findings suggest that AI-generated images have strong potential as
instructional media but require expert validation to prevent misconceptions.

Keywords: AI-Generated Images- GenAI in Education- Google Gemini

Topic: AI for Learning