Abstract
Canton embroidery is one of China's intangible cultural heritages, reflecting the "fusion of East and West" in the Lingnan region. However, it faces a crisis of transmission due to high production costs and a dwindling practitioner base. This paper explores a sustainable path for the craft by integrating generative AI into the design process. By combining Design Thinking with a workflow involving ChatGPT for prompt engineering and Midjourney for image generation, the researchers demonstrate how AI can lower the barrier to entry for traditional crafts and appeal to a younger, tech-savvy consumer base.
Introduction: The Transmission Crisis
Canton embroidery suffers from a significant gap between government preservation efforts and market reality. Traditionally, becoming a master requires over five years of training, leading to high production costs that restrict the art to high-end collectors. To ensure sustainable development, the craft must modernize its design phase—a traditional bottleneck—without losing its cultural essence. The challenge lies in making this heritage accessible to younger generations who prefer practical, affordable items over expensive wall hangings.
Methodology
The researchers utilized a Design Thinking framework, specifically the Double Diamond model, to create a four-stage functional model for AI-assisted embroidery design.
1. Foundation and Fieldwork
The team conducted research at the Guangzhou Embroidery Craft Factory, using a Deta digital colorimeter to extract precise color data from silk threads. They also analyzed 124 paintings by Qing Dynasty artist Ju Lian to establish a thematic baseline. This resulted in a custom digital library of Canton embroidery colors mapped to RGB and CMYK values.
2. The AI-Assisted Workflow
The core of the methodology involves a collaborative human-AI process. ChatGPT was employed for prompt engineering, translating artistic features (Theme, Color, Style, Composition, and Atmosphere) into structured inputs. These prompts were then fed into Midjourney to generate design sketches.
The team focused on a "Lotus and Butterfly" theme, iterating the process to select the most viable designs for physical production, such as image printing and embroidery weaving.
Key Insights and Results
The study revealed a stark "Age Acceptance Gap": 100% of respondents under 25 were willing to buy AI-influenced embroidery, compared to only 5.56% of those over 35. This suggests AI serves as a tool for market expansion rather than a replacement for the traditional collector market.
The model achieved a Customer Satisfaction Index (CSI) of 61.05, surpassing the "acceptable" threshold. Young consumers showed a clear preference for practical items like notebooks (20.5%) and accessories (18.41%) over traditional fine hangings (6.28%).
Discussion and Limitations
While AI excels at ideation and composition, the "soul" of the craft remains in the physical embroidery techniques that require human expertise. Current AI models still struggle with regional specificities, such as identifying local Lingnan flora and fauna, requiring manual intervention during prompt engineering. Furthermore, translating complex AI-generated images into physical stitches remains labor-intensive; designs that are too intricate may exceed the capacity of human artisans.
Conclusion
Integrating AI into the Canton embroidery design process offers a viable path for heritage preservation. By lowering the creation threshold and aligning with modern consumer preferences, AI can help transition traditional crafts from "exquisite collectibles" to "sustainable lifestyle products." Future research will focus on refining the prompt-to-stitch workflow to further bridge the gap between digital generation and physical craftsmanship.
Summary updated