Research on the Application of Song Dynasty Peony Patterns Based on the SD-LoRA Model

Abstract

 This paper responds to the growing use of AIGC in the cultural and creative industries. It targets two common issues in general-purpose foundation models for traditional pattern generation: distortion of cultural features and weak structural control. We propose an SD–LoRA modeling pathway that integrates design semiotics and cultural translation theory, and use Song-dynasty Luoyang peony motifs to explore a digital innovation route for traditional pattern design. First, we construct a high-quality peony motif dataset from authoritative museums in China. We then add semantic annotations across three translation layers: form, imagery, and context. Second, we combine LoRA fine-tuning with Stable Diffusion to train a LoRA model that can accurately capture and reproduce the aesthetic features of Song patterns. Finally, we enable controllable generation and practical application through parameter tuning. The proposed method supports the digital regeneration and creative transformation of Song peony motifs. The generated outcomes retain traditional spirit while presenting modern visual appeal. This work offers an efficient and culturally grounded paradigm for cultural product design.