Picture-Cued Writing Using AI-Generated Images for Language Acquisition

Abstract

The integration of AI-generated images into foreign language writing tasks has gained increasing attention in educational research. This study explores the impact of text-to-image generation models on students’ perceptions, revision behavior, and writing performance. We conducted a two-week experiment in a Japanese high school, where 34 students participated in picture-cued writing tasks facilitated by an AI system. Upon submission, students received AI-generated images based on their writing and automated feedback on their text, allowing them to independently decide whether to revise or finalize their writing. Findings indicate that while a majority of students improved their writing proficiency, most did not actively revise their work, and the proportion of lower-level writing increased over time. Despite positive perceptions of enjoyment and usefulness, students expressed neutral intentions toward future use, citing slow computation speed as a limitation. This study provides preliminary evidence that AI-generated images can support foreign language writing. The findings highlight the potential of multimodal AI tools in fostering linguistic accuracy in second language writing.