基于扩散模型微调范式的蜡染图案生成设计研究

汪鑫月; 吕健; 侯宇康; 周鑫; 林俊希

doi:10.19554/j.cnki.1001-3563.2026.02.021

PDF(30671 KB)

包装工程（设计栏目） ›› 2026, Vol. 47 ›› Issue (2) : 223-231. DOI: 10.19554/j.cnki.1001-3563.2026.02.021

视觉传达设计

基于扩散模型微调范式的蜡染图案生成设计研究

汪鑫月^a,b, 吕健^b,c,*, 侯宇康^c, 周鑫^b, 林俊希^b

作者信息 +

Wax Printing Pattern Generation Design Based on Diffusion Model Fine-tuning Paradigm

WANG Xinyue^a,b, LYU Jian^{b, c,*}, HOU Yukang^c, ZHOU Xin^b, LIN Junxi^b

Author information +

文章历史 +

摘要

目的为了探究人工智能视域下民族图案的智能设计方法和策略,以蜡染图案为例,提出了一种基于扩散模型微调范式的蜡染图案生成设计方法。方法首先,以小规模含有文本标注的实物蜡染样本数据集训练蜡染风格化LoRA模型;其次,以稳定扩散模型SDXL作为蜡染图案生成的基线模型,使用低秩自适应LoRA方式对SDXL注入可训练层进行微调训练;最后,为解决传统扩散模型生成图像的随机性,引入控制网络ControlNet对输入图像提取边缘特征信息,作为条件引导控制扩散模型生成个性化的蜡染图案。结果实验结果表明,能够通过文本条件或图像条件生成具有高真实感的蜡染风格创新图案,相较于SD1.5和CycleGAN两种图像生成模型,在FID和SSIM指标上均取得了更好的结果,验证了方法的可行性和有效性。结论以蜡染图案的智能创新设计为例,将生成式AI应用于传统民族图案领域,对民族图案的现代设计转化和发展具有重要意义。

Abstract

The work aims to propose a wax printing pattern generation design method based on diffusion model fine-tuning paradigm with the wax printing pattern as an example, to explore the intelligent innovative design methods and strategies of national patterns from the perspective of artificial intelligence. Firstly, the wax printing stylized LoRA model was trained with a small-scale physical wax printing sample data set with text annotation. Secondly, the stable diffusion model SDXL was used as the baseline model for wax printing pattern generation, and the SDXL was injected into the trainable layer for fine-tuning training using the low-rank adaptive LoRA method. Finally, in order to solve the randomness of the image generated by the traditional diffusion model, the control network ControlNet was introduced to extract the edge feature information of the input image as a condition to guide the control diffusion model to generate personalized wax printing patterns. The experimental results indicated that innovative wax printing patterns with high realism could be generated through text or image conditions. Compared with the two image generation models of SD1.5 and CycleGAN, it achieved better results in FID and SSIM indexes, which verified the feasibility and effectiveness of the method. Taking the intelligent innovative design of wax printing patterns as an example, the application of generative AI in the field of traditional ethnic patterns is of great significance to the transformation and development of modern design of ethnic patterns.

导出引用1

汪鑫月, 吕健, 侯宇康, 周鑫, 林俊希. 基于扩散模型微调范式的蜡染图案生成设计研究[J]. 包装工程. 2026, 47(2): 223-231 https://doi.org/10.19554/j.cnki.1001-3563.2026.02.021

WANG Xinyue, LYU Jian, HOU Yukang, ZHOU Xin, LIN Junxi. Wax Printing Pattern Generation Design Based on Diffusion Model Fine-tuning Paradigm[J]. Packaging Engineering. 2026, 47(2): 223-231 https://doi.org/10.19554/j.cnki.1001-3563.2026.02.021

中图分类号： TB472 TS941.26

参考文献

[1] 覃伟, 王宝升. 贵州靛青蜡染的产品创新设计策略[J]. 包装工程, 2018, 39(24): 276-282.
QIN W, WANG B S.Product Design Strategy Based on Indigo Batik of Guizhou[J]. Packaging Engineering, 2018, 39(24): 276-282.
[2] 岳迪, 吕健, 付倩文, 等. 基于可拓表征与神经网络的苗族图案设计[J]. 计算机系统应用, 2021, 30(3): 234-242.
YUE D, LYU J, FU Q W, et al.Miao Pattern Design Based on Extensional Representation and Neural Network[J]. Application of Computer Systems, 2021, 30(3): 234-242.
[3] 丁宁, 余隋怀, 初建杰, 等. 面向产品设计的民族图案语义量化模型构建与应用[J]. 计算机辅助设计与图形学学报, 2023, 35(4): 621-632.
DING N, YU S H, CHU J J, et al.Construction and Application of Ethnic Pattern Semantic Quantitative Model for Product Design[J]. Journal of Computer-Aided Design and Graphics, 2023, 35(4): 621-632.
[4] GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al.Generative Adversarial Networks[J]. Communications of the ACM, 2020, 63(11): 139-144.
[5] XUE A.End-to-End Chinese Landscape Painting Creation Using Generative Adversarial Networks[C]// 2021 IEEE Winter Conference on Applications of Computer Vision (WACV). Waikoloa: IEEE, 2021: 3863-3871.
[6] 冉二飞, 贾小军, 喻擎苍, 等. 基于SE注意力CycleGAN的蓝印花布单纹样自动生成[J]. 丝绸, 2024, 61(1): 31-37.
RAN E F, JIA X J, YU Q C, et al.Single Pattern Automatic Generation of Blue Calico Based on Se Attention CycleGAN[J]. Journal of Silk, 2024, 61(1): 31-37.
[7] Ho J, Jain A, Abbeel P.Denoising Diffusion Probabilistic Models[J]. Advances in Neural Information Processing Systems, 2020, 33: 6840-6851.
[8] 陈彦博, 王昱然, 李建亮. AIGC视域下的传统纹样创新设计研究——以宋代生色花为例[J]. 工业工程设计, 2024, 6(4): 63-71.
CHEN Y B, WANG Y R, LI J L.Innovative Design of Traditional Pattern from AlGC Perspective:Taking the Colored Flower Pattern of the Song Dynasty as an Example[J]. Industrial Engineering Design, 2024, 6(4): 63-71.
[9] HU E, SHEN Y, WALLIS P, et al. LoRA: Low-Rank Adaptation of Large Language Models[EB /OL]. (2021- 10-02) [2024-04-26]. https://arxiv.org/abs/2106.09685.
[10] 纪乐福, 王永江, 李启正. 基于DreamBooth的傣锦图案人工智能生成模型[J]. 服装学报, 2024, 9(5): 433-442.
JI L F, WANG Y J, LI Q Z.Artificial Intelligence Generation Model for Dai Brocade Pattern Based on DreamBooth[J]. Journal of Clothing Reserach, 2024, 9(5): 433-442.
[11] SHI Z, XIONG B.Fine-Tuning Text-to-Image Generation Models Using Curriculum Learning for Yao Costume Image Generation[C]// 2024 5th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT). Nanjing: IEEE, 2024: 1305-1311.
[12] HOULSBY N, GIURGIU A, JASTRZEBSKI S, et al.Parameter-Efficient Transfer Learning for NLP[C]// International Conference on Machine Learning, 2024
[13] LI X L, LIANG P.Prefix-Tuning: Optimizing Continuous Prompts for Generation[J]. Arxiv Preprint Arxiv: 2101.00190, 2021.
[14] 马诗洁, 徐华艺, 李聪聪, 等. 基于大模型微调范式的绘画风格模拟方法[J]. 计算机应用, 2024, 44(S1): 268-272.
MA S J, XU H Y, LI C C, et al.Painting Style Simulation Method Based on Fine-Tuning Paradigm for Large Models[J]. Computer applications, 2024, 44(S1): 268-272.
[15] 刘雨生, 肖学中. 基于扩散模型微调的高保真图像编辑[J]. 计算机应用, 2024, 44(11): 3574-3580.
LIU Y S, XIAO X Z.High-Fidelity Image Editing Based on Fine-Tuning of Diffusion Model[J]. Computer Applications, 2024, 44(11): 3574-3580.
[16] AUSTIN J, JOHNSON D D, HO J, et al.Structured Denoising Diffusion Models in Discrete State-Spaces[J]. Advances in Neural Information Processing Systems, 2021, 34: 17981-17993.
[17] 解晨辉, 李荣荣. 基于ControlNet的儿童床设计效果图生成技术[J]. 2024(2): 184-191.
XIE C H, LI R R.Children Bed Design Effect Diagram Based on Controlnet[J]. Journal of Forestry Engineering, 2024(2): 184-191.
[18] 涂晴昊, 李元琪, 刘一凡, 等. 基于扩散模型的文本生成材质贴图的泛化性优化方法[J]. 图学学报, 2025, 46(1): 139-149.
TU Q H, LI Y Q, LIU Y F, et al.Generalization Optimization Method for Text to Material Texture Maps Based on Diffusion Model[J]. Journal of Graphic Science, 2025, 46(1): 139-149.
[19] LI C F, BOVIK A C.Content-Partitioned Structural Similarity Index for Image Quality Assessment[J]. Signal Processing: Image Communication, 2010, 25(7): 517-526.