FLUX–CPSC: a multimodal-driven framework for high-fidelity transfer of traditional Chinese opera costume elements
Yuxuan Guo, Kaixuan Liu · Digital Scholarship in the Humanities · 2026
Abstract To promote the modern inheritance and fashion-oriented transformation of traditional Chinese opera costume culture, and address the core issue that existing style transfer techniques fail to simultaneously ensure cultural symbol integrity, costume structural stability, and embroidery craftsmanship fidelity, this study proposes a high-fidelity transfer method for opera costume elements for modern fashion design. The FLUX–Cross-modal Patch-level Style Control (FLUX–CPSC) paradigm is proposed, which includes two core modules: a multimodal hierarchical fusion module and an opera costume-specific Cross-modal Patch-level Style Control–Low-Rank Adaptation (CPSC–LoRA) semantic calibration module. It integrates FLUX-Redux with Segment Anything Model-based segmentation for interference-free core style feature extraction, uses DepthAnythingV2 to construct spatial constraints and obtain modern costume depth features, dynamically allocates text, style, and depth feature fusion weights by constraint priority, generates preliminary transfer images via FLUX sampling, and adopts pretrained CPSC–LoRA for targeted pattern semantic deviation calibration. Experimental results demonstrate that FLUX–CPSC outperforms mainstream methods in both objective metrics and subjective evaluations, and exhibits superior performance in pattern recognizability, color fidelity, and structural retention. This verifies its effectiveness in the high-fidelity transfer of opera costume cultural symbols. Proposing a constraint priority-based multimodal fusion mechanism to resolve feature interference in high-semantic-density tasks. Designing a lightweight CPSC–LoRA module for accurate semantic mapping of opera costume patterns. Integrating FLUX-Redux to realize high-fidelity style transfer under few-shot conditions. It provides a reusable technical paradigm for the digital inheritance of opera costume culture, bridges Chinese opera costume heritage and AI-assisted design, and promotes digital innovation and inheritance of traditional costume elements.