<p>Art decorative pattern design needs to satisfy both the uniqueness of style and the rigor of composition. Traditional methods are difficult to achieve efficient and controllable generation. To build a generation framework that can accurately and collaboratively control pattern style, layout and semantics, an art decorative pattern design method based on a multi-modal structure semantics-enhanced stable diffusion model is proposed. A double-layer U-Net structure is designed to decouple layout and style control, and combine low-rank adaptation and block-level fine-grained fine-tuning for efficient parameter optimization. A prompt system of multi-modal layered conditional constraints and semantic deconstruction is introduced to optimize semantic guidance. Experiments showed that this method led in many key indicators, with its style consistency score, layout alignment and text-image similarity reaching 0.891, 0.863 and 0.310, respectively. Under complex composition, its structural similarity index and peak signal-to-noise ratio reached 0.785 and 24.891&#xa0;dB, respectively. Ablation experiments and cross-type pattern generation tests further verified the effectiveness of each module and the generalization ability of the model. The results show that the integrated method effectively solves the collaborative problems of style fidelity, spatial controllability and semantic alignment in generating artistic decorative patterns, and provides a new paradigm for intelligent design.</p>