从特征辨识到图像生成:基于AIGC范式的苗族服饰设计OA北大核心CSTPCD
From feature recognition to image generation:Miao ethnic costume design based on the AIGC paradigm
苗族服饰是中国民族服饰体系中重要的组成部分,在商品化发展中开辟了广阔的市场,但产品同质化现象将阻碍民族文化的进一步发展,人工智能生成内容(AIGC)的发展为传统服饰设计提供了新的可能.文章基于苗族服饰现代化的需要,通过搭建多模型组合流程,为苗族服饰成衣化设计提供灵感样本.实验结果表明,以服饰辨识性为基础,采用Lora训练方式,仅需少量样本即可实现苗族服饰特征在Stable Diffusion生成内容中的迁移.将深度学习引入非遗服饰现代化的进程中,旨在助力非遗服饰朝着成衣化、多样化、大众化方向发展,以数字化方式推动民族服饰转型升级.
Miao ethnic costumes constitute a vital component of China's ethnic costume system,carving out a significant market presence in the realm of commercial development.These costumes play a crucial role in cultural and tourism activities,with their distinctive characteristics not only fulfilling the yearnings of tourists for exotic experiences but also elevating the influence of Miao culture.However,the market is inundated with a plethora of homogeneous products.Ordinary designers may struggle to comprehend the profoundness of Miao culture,and inheritors of intangible cultural heritage find it challenging to balance the cultural preservation aspect with the commercial viability of garment design.The phenomenon of product homogenization poses a hindrance to the continued development of ethnic culture.The question of how to enrich Miao costume design and consistently offer diverse Miao costume products becomes a topic that warrants careful consideration. The emergence of the Latent Diffusion Model(LDM)has opened up new possibilities for traditional costume design.Based on the modernization needs of Miao ethnic costumes,the Latent Diffusion Model SD1.5 served as the foundational model.Analyzing identifiable features in Miao ethnic costumes from the perspectives of style,accessories,patterns,and color,and with them as criteria for selecting training set images,the study applied the Low-Rank Adaptation of Large Language Model(Lora)method to construct a few-sample style transfer model.The ultimate control of generated content was achieved through the use of Control Net.By establishing a multi-model combination process,inspirational samples were provided for the ready-to-wear design of Miao ethnic costumes.Experimental results indicate that,based on costume recognizability and using the Lora training method,only a small number of samples are needed to achieve the transfer of Miao ethnic costume features in the content generated by Stable Diffusion.The few-sample Miao style transfer model based on the recognizability of Miao ethnic costumes possesses certain advantages in reducing the amount of training materials,minimizing training steps,and shortening training time.It plays a positive role in enriching the ready-to-wear design of Miao intangible cultural heritage costumes. By introducing deep learning into the modernization of intangible cultural heritage costumes,the study aims to revitalize ethnic costumes through digital means.The application of Artificial Intelligence Generated Content(AIGC)is intended to provide more design inspiration for traditional intangible cultural heritage costume design,supporting its development towards ready-to-wear,diversification,and popularization.Intangible cultural heritage costumes can not only better integrate into modern fashion trends but also achieve diversity through innovative design to meet the needs of different groups.The digital approach not only improves the efficiency of costume design,but also injects new inspiration and elements into traditional craftsmanship,promoting the transformation and upgrading of ethnic costumes through digital means.
于鹏;张毅
江南大学 设计学院,江苏 无锡 214122
轻工业
AIGC苗族非遗服饰少样本风格迁移Stable DiffusionLora辅助设计
AIGCMiao ethnic groupintangible cultural heritage costumesfew-sample style transferStable DiffusionLoraassisted design
《丝绸》 2024 (003)
1-10 / 10
教育部人文社会科学研究一般项目(21YJA760096);中国非物质文化遗产传承人研修培训计划项目(文非遗发[2017]2号)
评论