• 목록

Generative AI image tools such as Midjourney have rapidly transformed the digital content landscape, making it increasingly difficult to distinguish AI-generated images from real ones. While previous studies have focused on detection performance or technical evaluation, systematic classification of visual error patterns from a design perspective remains limited. This study conducts a case study analysis of 12 Midjourney-generated images (v8.1) across four content types - portrait, text scene, composite object, and landscape - and classifies the visual errors observed into five cue categories: structural, surface, optical, semantic, and stylistic. The findings propose a visual error taxonomy that can serve as a foundational framework for AI content literacy education and the improvement of generative AI tools.