Revisiting the Generation Framework in Generative AI-Based Intelligent Structural Design

Generative AI is reshaping structural design; however, current generative AI-based methods primarily focus on enhancing the AI models employed, overlooking another critical factor that influences outcomes: the generation framework. Existing frameworks fall into two categories: the end-to-end generation framework incorporating generative AI (EGAI), which includes EGAI-GAN and EGAI-DM, and is widely adopted by nearly all current methods; and the two-stage GAI (TGAI), specifically TGAI-DM + PCDM, proposed by generative AIBIM—one of the most representative pipelines. Generative AIBIM demonstrates, from a theoretical standpoint, that TGAI is more effective than EGAI; however, it lacks empirical validation. This paper designs EGAI-DM + PCDM and conducts a comparative analysis with TGAI-DM + PCDM. The experimental results indicate TGAI-DM + PCDM enables AI models to generate design drawings with greater precision and enhanced perceptual quality compared to EGAI-DM + PCDM. These findings substantiate that TGAI is indeed more effective than EGAI. We believe this conclusion will inspire further exploration and broader applications of TGAI.