LoRA模型微调Stable Diffusion的设计师风格服装生成方法

Designer-style Clothing Generation Method Based on LoRA Model Fine-tuning Stable Diffusion

为控制Stable Diffusion模型准确生成特定设计师风格的服装图像,应用LoRA模型微调Stable Diffusion生成过程的方法实现风格化服装图像生成。以中国十佳时装服装设计师孙林的设计风格为例,首先对设计师的发布会服装进行系统分析,提取设计师的风格特征因子,根据设计师风格因子分别训练3个风格化LoRA模型,分别验证每个风格因子的表达效果。最后,通过3个LoRA模型叠加使用的方式,生成具有设计师风格特征的服装设计图像。结果表明:同时调用多个风格化LoRA模型微调Stable Diffusion的生成过程,能够使生成的服装符合设计师的服装风格,实现设计师风格的迁移设计,提高设计师进行服装系列设计和产品开发的效率。

In order to control the Stable Diffusion model accurately generating clothing images in a specific designer's style,the LoRA model is applied to fine-tune the Stable Diffusion generation process to achieve stylized clothing image generation.Taking the design style of Sun Lin,one of China's top ten fashion designers,as an example,firstly,the clothing from the designer's fashion show is systematically analyzed to extract the designer's stylistic features.Then three stylized LoRA models are crafted according to the designer's style factors,and the effectiveness of each style factor is individually assessed.Finally,by simultaneously invoking the three LoRA models,clothing images with designer-style characteristics are generated.The experimental results show that invoking multiple stylized LoRA models to fine-tune the Stable Diffusion generation process can make the generated clothing images conform to the desired designer's style,realize the transfer design of designer style,and improve the efficiency of designers in clothing series design and product development.