In the context of computational textile design and intelligent manufacturing, the key challenge lies in collaboratively achieving both the functionality and artistic expression of fabrics. Existing methods often focus on a single dimension and lack a unified framework that integrates multi-physics constraints with artistic semantics. To address this, this paper proposes a generative artificial intelligence-based method, with the core innovations being: 1) a dual-path decoupled representation of yarn arrangement and weave patterns; 2) the design of a function-semantic dual encoding and joint generation framework; 3) the introduction of curriculum learning strategies to enhance training stability. Experiments are based on a self-built dataset, TextileArt-Weave v1.0 (8,642 samples), covering three types of functional fabrics and five artistic styles. Quantitative results show that the proposed method significantly outperforms visual generation and engineering CAD baselines in functional indicators (porosity error: 3.2 ± 0.4% vs. 9.8 ± 0.6% / 12.4 ± 0.8%), and outperforms rule-driven methods in artistic semantic alignment (CLIP Score: 0.78 ± 0.02 vs. 0.32 ± 0.02), while maintaining excellent manufacturability (interweaving cycle compliance: 96.8 ± 0.7%, whereas pure visual models yield 0%). Ablation studies and robustness validation show that each module makes a clear contribution to functionality, artistry, and manufacturability, and the method remains stable under noise, material changes, and process disturbances. To the best of our knowledge, this study presents one of the first end-to-end frameworks for mapping multimodal prompts to manufacturable fabric structures under joint functional and artistic constraints and has open-sourced the textile multimodal benchmark dataset, providing feasible paths for smart wearables, cultural digitization, and sustainable fashion.

generative artificial intelligence, functional textile structures, artistic expression, multimodal generation, computational textile design