ICHGen: a framework for cloud-collar generation and personalised virtual try-on

Purpose This study proposes ICHGen, a framework for cloud-collar generation and personalised virtual try-on. It aims to integrate traditional structural knowledge and physically informed wearing priors into generative artificial intelligence, supporting culturally consistent design, efficient visualisation and the digital reuse of intangible cultural heritage. Design/methodology/approach ICHGen integrates Shape Grammar, FLUX-LoRA, Position-Based Dynamics (PBD), and Kontext. SG-derived structural guidance is introduced into cloud-collar generation, while PBD-simulated wearing samples provide physical priors for Kontext training. The framework is evaluated using geometric metrics, expert assessment, baseline comparisons and PBD-reference consistency analysis. Findings ICHGen achieved the highest expert Shape grammar Compliance Score (9.51 ± 0.10) among the evaluated generation methods and an overall VTON score of 4.33 ± 0.17. PBD-reference evaluation also showed substantial consistency in garment region and occlusion, while the complete generation-and-try-on process required approximately 18 s per case. Practical implications The framework provides a practical workflow for cloud-collar digital design, personalised virtual presentation, and creative reuse. By incorporating structural and physical priors during model construction, it supports efficient image-based generation without requiring real-time physical simulation. Originality/value The cloud-collar case demonstrates a practical approach to integrating explicit cultural and physical knowledge into generative AI for digital heritage applications. It lowers the threshold of cloud collar design, promotes public participation in intangible cultural heritage innovation, providing a reusable technical reference for the digital revitalisation and intelligent manufacturing of ICH.