Scholarlink

<p>This study examines the potential and limitations of Artificial Intelligence Generated Content (AIGC) in reconstructing Qing imperial interiors, with <i>Juanqinzhai</i> as a case study. A high-fidelity SketchUp(SU) reference model and historical archives served as the ground truth for both geometric and semantic validation. Over 200 images were generated via function-oriented prompts using Midjourney v6 and Stable Diffusion XL. Results show systematic errors (6.5–40.6%), including exaggerated spatial depth, disproportionate partitions, and undersized ceiling elements, revealing a bias toward visual spectacle rather than structural fidelity. Semantic analysis further identified ornamental exaggeration and stylistic hybridization, reflecting cross-cultural biases embedded in training data. A three-stage critical generation workflow is proposed, combining AIGC’s creative diversity with expert correction and Historic Building Information Modeling (HBIM) integration. The findings underscore the risks of uncritical use of generative tools while demonstrating their value as auxiliary methods for culturally sensitive digital heritage reconstruction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A critical Artificial Intelligence-generated content approach for the reconstruction of Qing Palace interiors: the case of Juanqinzhai

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月27日

查看全文
  • Changqing Wei,
  • Jiaru Liu,
  • Jing Jia,
  • Dongyi Kong,
  • Meng Yuan,
  • Siyu Yan

摘要

This study examines the potential and limitations of Artificial Intelligence Generated Content (AIGC) in reconstructing Qing imperial interiors, with Juanqinzhai as a case study. A high-fidelity SketchUp(SU) reference model and historical archives served as the ground truth for both geometric and semantic validation. Over 200 images were generated via function-oriented prompts using Midjourney v6 and Stable Diffusion XL. Results show systematic errors (6.5–40.6%), including exaggerated spatial depth, disproportionate partitions, and undersized ceiling elements, revealing a bias toward visual spectacle rather than structural fidelity. Semantic analysis further identified ornamental exaggeration and stylistic hybridization, reflecting cross-cultural biases embedded in training data. A three-stage critical generation workflow is proposed, combining AIGC’s creative diversity with expert correction and Historic Building Information Modeling (HBIM) integration. The findings underscore the risks of uncritical use of generative tools while demonstrating their value as auxiliary methods for culturally sensitive digital heritage reconstruction.