Multimodal generative AI product design based on cultural philosophy and quantitative constraints

Generative artificial intelligence (GAI) in product design frequently encounters a ‘translation gap’ between abstract cultural philosophy and engineering specifications, alongside ‘semantic drift’ across multimodal generation chains. To address these challenges, this paper presents a General Protocol for Cultural-Philosophy-Driven Generative Design, demonstrated through the development of a ‘Zhonghe’ (Harmonious Moderation) Ming-style armchair. The methodology formalises cultural knowledge into a machine-executable Form–Meaning–Technique (FMT) parameter dictionary, distilled from a classical corpus of 1,287,605 characters. By integrating Stable Diffusion with ControlNet and a projection-based 3D verification loop, the workflow enforces cross-modal consistency through three quantitative metrics: bilateral symmetry (S), motif matching (M), and colour difference (ΔE via CIE76). Validation thresholds are statistically determined using the ROC–Youden index (S≥0.80,M≥0.85,ΔE<5) to trigger targeted backtracking upon violation. Comparative results indicate that this closed-loop protocol significantly outperforms unconstrained baselines (e.g. SD-only, MJ-only), achieving high-fidelity outcomes (Best Scheme: S ≈0.968,M≈92.6\%,ΔE≈2.11) while ensuring structural feasibility. This study contributes an auditable, quantifiable decision framework that bridges the disconnect between traditional cultural aesthetics and modern engineering design constraints.