Knowledge Base Repository
In addition to research papers, the Design Society is developing several valuable resources for those interested in the study of design. These include a repository of PhD theses, a library of case studies and transcripts of design activities, and an archive of our newsletters. Please note that these resources are accessible exclusively to Design Society members.
KNOWLEDGE-AUGMENTED AGENTIC SYSTEMS FOR TRADITIONAL COSTUME PATTERN DESIGN
Jia SHANG; Yihui YAO; Muchen ZHOU
Type:
Year:
2025
Editor:
Yong Se Kim; Yutaka Nomaguchi; Cees de Bont; Jianxi Luo; Xiaofang Yuan; Linna Hu; Meng Wang
Author:
Series:
Other endorsed
Institution:
Shanghai University, China
Page(s):
106-114
Abstract:
Generative AI has accelerated visual design workflows, yet it remains limited in domains that require cultural sensitivity, compositional precision, and production feasibility. Traditional costume pattern design exemplifies these challenges, demanding alignment with symbolic meaning, design conventions, and craft constraints. We present a knowledge-augmented agentic system tailored for professional design tasks. Powered by a large language model, the system follows a closed loop of perception, retrieval, generation, reflection, and revision. It integrates tools such as pattern-aware foreground segmentation, motif-level inpainting, symmetry-based layout editing, and style-guided color adjustment to support culturally and structurally grounded outputs. The system leverages a structured retrieval module, including 504 design rules, 781 expert cases, and a curated style bank of 20 annotated reference images. Retrieved knowledge is translated into actionable design constraints, guiding iterative refinement throughout the process. This approach enables deeper integration of generative AI with traditional design practice, offering a system capable of producing culturally coherent, visually balanced, and production-ready costume patterns. The approach provides a viable pathway for integrating cultural context, aesthetic principles, and craft constraints into generative workflows.
Keywords:
