생성형 AI 이미지 기반 영상용 실내 장면 공간 구조화 연구

초록

This study explores a method for structuring a single generative-AI-based indoor scene image into a space suitable for video production. Generative AI can rapidly create visually persuasive scene images from text and image inputs, but the completeness of a single image does not necessarily establish a usable scene space. In video, a scene is divided into multiple shots, and spatial continuity depends on maintaining camera direction, character movement, object placement, and spatial anchors. However, when viewpoints change or unseen areas are generated, AI-generated images tend to reconstruct visually plausible spaces rather than preserve existing spatial relationships. Therefore, this study treats generative AI images not as final outputs but as intermediate design drafts for examining and structuring indoor scene spaces. The study involved reference scene selection, multi-view generation, scene floor plan construction, six-direction viewpoint review, spatial expansion, and shot selection. Through this process, it analyzed maintained and reinterpreted elements and identified points requiring creator intervention. The results show that AI-generated images generally preserved visual impressions such as color, lighting, and atmosphere, while object placement, directionality, and shot continuity became unstable during viewpoint changes and spatial expansion. The scene floor plan and six-direction viewpoint review provided criteria for identifying these inconsistencies and structuring images for shot composition. These limitations are interpreted not simply as generation errors but as production conditions requiring the examination of camera, movement, directionality, and shot continuity. Accordingly, this study proposes a structuring process and methodological criteria for converting a single generated image into an indoor scene space applicable to shot composition.

키워드