Exploring dual pathways for traditional pattern innovation: shape grammar and diffusion models

Traditional decorative motifs embody rich cultural heritage, yet innovating such patterns without eroding their essence remains challenging. This study systematically compares rule-based Shape Grammar and AI-driven Denoising Diffusion Probabilistic Models (DDPM) for “Baoxiang” pattern design. We assembled a dataset of Baoxiang images via Python scripts from online galleries and museum collections, then employed the Segment Anything Model (SAM) to isolate key motif elements. Both Shape Grammar and DDPM were used to generate novel designs, which were evaluated through a public survey assessing innovation, aesthetics, complexity, and cultural fidelity. ANOVA confirmed that Shape Grammar better preserves traditional structure, whereas DDPM delivers greater creative diversity. These findings highlight complementary strengths of each approach and inform future hybrid methodologies for culturally respectful pattern innovation.