Scholarlink

<p>The Miao ethnic batik technique, as a State-level intangible cultural heritage, holds great significance for both inheritance and innovation. At present, it is impossible to meet the requirements of efficient and diverse batik pattern design by relying solely on manual labor, while the traditional computer-aided pattern design also suffers from these issues, such as insufficient cultural relevance, inappropriate artistic expression and deficient innovativeness. In this paper, an automatic generation method of the Miao batik patterns based on the improved Stable Diffusion model is proposed, namely an image dataset containing the Miao batik patterns and the corresponding text descriptions are constructed, which is also one of the earlier text-to-image datasets related to the Miao batik patterns at present. Using the Stable Diffusion model based on deep learning, the generation model for the pattern design is established. To capture the abstract style of the Miao batik, the semantic information from the Stable Diffusion model is integrated with the LoRA module which carries the Miao ethnic batik style. Concurrently, the network structure of the model is also modified to enrich the features of the batik images, which ensures the correlation between the automatically generated images and linguistic description, and the Miao culture elements, and accurately reflects the stylistic characteristics of the Miao batik patterns as well. The generated patterns have achieved the positive results in terms of the evaluation of techniques, such as FID and IS, indicating that the generated images have greater diversity while ensuring semantics. In the aspect of artistic evaluation, the comprehensive scores of both professional and non-professional groups on the generated patterns are all above 3.8 points (out of 5 points), which suggests that the generated patterns can satisfy the design requirements and practical applications to a great extent. The limitation of this study lies in the single colors of the batik patterns, which restricts the design expansibility of the model, so the subsequent experiments can be conducted in combination with multi-color patterns to enhance the adaptability of the generated model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic generation and design of the Miao batik patterns based on the stable diffusion model

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Tianyuan Guo,
  • Longguo Tian,
  • Quan Ning

摘要

The Miao ethnic batik technique, as a State-level intangible cultural heritage, holds great significance for both inheritance and innovation. At present, it is impossible to meet the requirements of efficient and diverse batik pattern design by relying solely on manual labor, while the traditional computer-aided pattern design also suffers from these issues, such as insufficient cultural relevance, inappropriate artistic expression and deficient innovativeness. In this paper, an automatic generation method of the Miao batik patterns based on the improved Stable Diffusion model is proposed, namely an image dataset containing the Miao batik patterns and the corresponding text descriptions are constructed, which is also one of the earlier text-to-image datasets related to the Miao batik patterns at present. Using the Stable Diffusion model based on deep learning, the generation model for the pattern design is established. To capture the abstract style of the Miao batik, the semantic information from the Stable Diffusion model is integrated with the LoRA module which carries the Miao ethnic batik style. Concurrently, the network structure of the model is also modified to enrich the features of the batik images, which ensures the correlation between the automatically generated images and linguistic description, and the Miao culture elements, and accurately reflects the stylistic characteristics of the Miao batik patterns as well. The generated patterns have achieved the positive results in terms of the evaluation of techniques, such as FID and IS, indicating that the generated images have greater diversity while ensuring semantics. In the aspect of artistic evaluation, the comprehensive scores of both professional and non-professional groups on the generated patterns are all above 3.8 points (out of 5 points), which suggests that the generated patterns can satisfy the design requirements and practical applications to a great extent. The limitation of this study lies in the single colors of the batik patterns, which restricts the design expansibility of the model, so the subsequent experiments can be conducted in combination with multi-color patterns to enhance the adaptability of the generated model.