AIFiligree: A Generative AI Framework for Designing Exquisite Filigree Artworks
Authors
Research Background and Problem
-
What problems or challenges did the authors identify?
The authors identified that while China's traditional filigree craftsmanship possesses significant aesthetic and cultural value, its complex design and production processes, requiring extensive time and specialized skills, limit its design aesthetics and cultural inheritance. Furthermore, although existing digital technologies have provided some assistance in integrating craftsmanship, most interventions result in outputs that are not entirely traditional, failing to address the intricate details of filigree craftsmanship. -
Why is this problem important?
Filigree craftsmanship is a representative of China's national intangible cultural heritage, holding great significance for cultural preservation and artistic innovation. However, due to its intricate production process and high technical requirements, many designers and craft enthusiasts face technical barriers, making participation difficult. Addressing this issue not only helps preserve traditional culture but also promotes cultural innovation. -
Research Motivation and Related Work
To address these limitations, the authors propose leveraging generative artificial intelligence (AIGC) technology to accelerate the design process of complex craftsmanship, thereby expanding the application scope of traditional crafts. Existing research primarily focuses on using digital technologies to lower the barriers to entry for craftsmanship and improving traditional craft design through AI, but it has not adequately resolved the detailed representation of filigree craftsmanship.
Solution
-
What methods or solutions did the authors propose?
The authors proposed a generative AI-based framework called AIFiligree. This framework integrates the Stable Diffusion model and the LoRA model to generate highly realistic and intricate filigree craftsmanship designs. It supports various design scenarios, including text-to-image (T2I) generation, image-to-image transformation (I2I), and localized inpainting. Additionally, the framework provides a library of traditional filigree craftsmanship prompts to enhance user experience. -
What are the innovative aspects of this solution?
- Utilizing AIGC technology to handle intricate craft details, making the design process more efficient.
- Offering a user-friendly design tool interface that supports natural language input and critical parameter adjustments.
- Optimizing labels and datasets specifically for the unique complexity of filigree craftsmanship.
- Training the LoRA model with an adaptive optimizer to improve the precision of image detail generation and cultural reproduction.
-
What are the implementation steps and key technologies used?
- Data Collection and Processing: Collected approximately 400 images for training, including filigree craft products and their descriptive texts. Unclear images were removed, and resolutions were standardized to 1024×1024 pixels.
- Label Optimization: Manually optimized image labels to ensure accurate cultural representation. Experiments showed that combining Chinese and English keywords yielded the best results.
- Model Training and Optimization: Conducted multi-round iterative training using adaptive optimizers (e.g., DAdaptLion) and compared the impact of different training steps on model performance.
- User Interface Development: Built a visual design workflow based on ComfyUI, supporting text input generation, multi-image references, and localized detail inpainting.
Research Outcomes
-
What specific outcomes were achieved?
- Successfully developed a LoRA model capable of generating high-quality filigree craft images and applied it to user design workflows through the AIFiligree tool.
- User experiments demonstrated that the tool significantly enhanced designers' efficiency and creativity while helping users better understand traditional culture and artistic expression.
- Expert users noted that the AI-generated images had high visual appeal and provided valuable cultural elements as references for designers.
-
What advantages does it have compared to existing solutions?
- Preserves the aesthetic features of traditional filigree craftsmanship while enabling innovative designs.
- Features a user-friendly interface with a low entry barrier, allowing non-expert users to participate in complex craft design.
- Supports the generation of highly detailed and culturally accurate traditional filigree works, addressing the shortcomings of existing digital tools in producing traditional-style outputs.
-
What are the experimental or evaluation results?
- In a study involving 22 filigree experts, AI-generated images scored similarly to traditional images in aesthetic appeal, with AI-generated works showing an advantage in creative expression.
- The system's usability test yielded an average score of 78.4 (SUS score), indicating user satisfaction with the tool's usability and learning curve.
- Users averaged 29 iterations, significantly exceeding the number of iterations in traditional design processes, demonstrating that AI effectively reduces design time investment.
-
Limitations and Future Directions
- Limitations:
- The diversity of image styles in the database is limited, affecting the ability to generate modern design styles.
- While AI can generate design effects, it cannot fully replicate the spirit and detailed expression of manual craftsmanship, and some generation errors persist.
- Future Directions:
- Expand the database to include a wider variety of design styles and filigree craft patterns.
- Explore integrating AIFiligree with actual production processes, such as using CAD and 3D printing technologies for rapid prototyping.
- Further optimize AI models to better learn the complex emotions and spiritual expressions of traditional craftsmanship, enhancing generation quality.
- Limitations:
Through this research, AIFiligree not only advances the digitization and innovation of traditional filigree craftsmanship but also provides a new solution for modernizing the design of traditional culture.
Research Questions / Practical Problems
Question signals indexed for this paper.
Research Questions
3- How can generative AI technologies handle complex details in traditional filigree craft design?Category: Material, Craft, and Fabrication-Driven Design ResearchSimilar questionsarrow_forward
- Can generative AI technologies improve designers' efficiency and creativity in filigree craft design?Category: Material, Craft, and Fabrication-Driven Design ResearchSimilar questionsarrow_forward
- How does the AIFiligree framework perform in cultural representation and design innovation?Category: Material, Craft, and Fabrication-Driven Design ResearchSimilar questionsarrow_forward
Practical Problems
1- Designers and non-experts struggle to participate in complex traditional filigree craft design.Category: Material, Craft, and Fabrication-Driven Design ResearchSimilar questionsarrow_forward
Based on Jaccard similarity of research subtopics & professions (≥60%)