표준화 프롬프트 조건에서 생성형 AI를 활용한 중국 전통정원 차경(借景) 의경 요소의 이미지화 연구 - 빛 환경 표현을 중심으로 -

초록

(Background and Purpose) Advances in artificial intelligence-generated content have expanded the use of text-to-image models in design visualization, landscape conception, and cultural heritage representation. Borrowed scenery in traditional gardens does more than place a distant view within an image; it creates an aesthetic experience through interior-exterior relationships, spatial layers, viewing position, and the light environment. This study aims to analyze how generative AI visualizes the artistic-conception elements of borrowed scenery in traditional Chinese gardens under standardized prompt conditions, to clarify the effects of light-environment descriptions on the visual expression of borrowed-scenery space and atmosphere, and to identify implications for the early stages of contemporary garden design. (Method) SDXL 1.0 was selected because its parameters can be recorded and its generation conditions controlled. Five experimental groups were established: basic borrowed scenery, spatial layers, soft morning side light, warm sunset backlight, and overcast diffuse light. The garden subject, distant-view object, viewing angle, and image style were held constant across groups. Image size, sampling method, generation parameters, and paired random seeds were also standardized. Twenty images were generated for each group, producing 100 experimental images. Two coders recorded major garden elements and relevant spatial and light-environment characteristics. Seven evaluators with backgrounds in garden, landscape, or environmental design then independently assessed the images on a five-point scale for distant-view incorporation, interior-exterior relationship, spatial layers, light-environment expression, and artistic-conception atmosphere. Inter-rater consistency tests and between-group comparisons were conducted. (Results) Distant mountains, water surfaces, plants, and traditional buildings were comparatively easy to visualize, whereas distant views generated with the basic prompt tended to remain ordinary background scenery. After foreground, middle ground, background, and interior-exterior connections were specified, the interior-exterior relationship score increased from 2.49 to 3.49, and the spatial-layers score rose from 2.74 to 3.86. This indicates that explicit spatial vocabulary strengthens the structure and visual depth of borrowed scenery. With light-environment descriptions added, scores ranged from 3.84 to 4.23 for light-environment expression and from 3.84 to 4.21 for artistic-conception atmosphere. Soft morning side light mainly produced calmness and clarity; warm sunset backlight enhanced a dark-near and bright-far relationship, spatial depth, and transitions between solidity and void; overcast diffuse light increased calmness and haziness but sometimes reduced distant-view clarity. The warm sunset backlight group achieved the strongest overall performance. (Conclusions) AIGC can help identify visual elements of borrowed-scenery artistic conception that can be verbalized and visualized. It may support the examination of borrowed-scenery relationships, comparison of light-environment alternatives, and communication of concepts during the early stages of contemporary garden design. Nevertheless, generated images remain static and cannot fully represent actual sites, spatial scale, viewing position, temporal change, bodily movement, or cultural association.This study combined AIGC image experiments with field investigation, sunlight analysis, spatial simulation, and viewer evaluation, thereby further verifying the relationship between image-based representation and real garden experience.

키워드