Generating Chinese intangible cultural heritage images with structure and color awareness

The visual identity of Chinese intangible cultural heritage (ICH) relies on structured composition and symbolic color traditions. We present Colorful Heritage, a structure- and color-aware generative framework for stylizing ICH visual forms with cultural fidelity. Our method integrates: (1) a structure-guided stylization module encoding principles such as axial symmetry; (2) a color-aware embedding mechanism for symbolic palette modeling; and (3) a disentangled two-stage training strategy to separate content from style. To support evaluation and future research, we introduce ICHStyleBench, a benchmark of 10 ICH styles—including paper cutting, clay sculpture, and embroidery—annotated with layout and color attributes. Both quantitative analysis and expert evaluations confirm that our method achieves superior performance in semantic integrity, visual symbolism, and compositional structure preservation, with scores of 84.6 in content preservation, 81.2 in style alignment, 0.502 in CLIP-S, and a user preference rate of 34.5%, significantly higher than all baselines. This work provides a computational contribution to the digital preservation and revitalization of intangible cultural heritage.