Research on Multi Style Transfer and Creative Assistance of Traditional Chinese Painting with Fusion Diffusion Model

Abstract

Traditional Chinese painting has a long history, and the inheritance and innovation of its diverse styles remain important in digital cultural communication. With the development of artificial intelligence, deep-learning-based style transfer has become a useful tool for digital dissemination and creative assistance. However, traditional painting style transfer still suffers from limited style restoration, poor flexibility in switching among multiple styles, insufficient creative support, and weak extraction of key artistic features. To address these problems, this study constructs a multistyle transfer and creative-assistance framework based on a fusion diffusion model. A multi-style dataset of traditional Chinese painting is first built, covering ink wash, meticulous, blue-green, and light red-ochre painting styles, with annotations for brushwork, color, composition, and artistic conception. Based on Stable Diffusion v1.5, a multi-scale feature extraction network with attention mechanisms is introduced to capture brush- and-ink texture, blank-space structure, and global style information. A multi-style transfer module and creative-assistance module are then designed for style switching, detail completion, sketch optimization, and style recommendation. Experiments show that style transfer accuracy reaches 92.6%, improving over GAN-based and baseline diffusion models, while inference speed improves through HiGS sampling. The image-generation framework also provides transferable value for electromagnetic visualization, where complex field patterns and wave propagation images require interpretable visual transformation.