To fully integrate key features of the target style in clothing design and effectively control the degree of style transfer, a method for generating artistic patterns for clothing design was designed. Addressing the limited data set of original clothing design patterns, various data augmentation methods, such as translation and scaling, were used to expand the dataset from 2,500 images to 12,500. “Aesthetic fit” and “style identity” were selected as key indicators, and a fourth-order evaluation scale was used to establish a fitness function. An adaptive style transfer model was constructed using the convolutional layers of a ResNet-34 network. Two key loss functions, structural loss and fractal loss, were set, and fractal features of the pattern were extracted using a total loss function. Pattern generation was performed by integrating pattern genes, including abstracting fractal elements from the original clothing pattern, depicting complex structures using an improved fractal function, and outputting the pattern genes as vectors. The application of fractal iteration and fusion techniques in pattern creation was also introduced, providing new insights into the generation of artistic patterns for clothing design. The experiment shows that the fusion rate can effectively regulate the degree of style conversion, and when the fusion is between 40% and 100%, the Baroque style gradually becomes stronger in tie dye patterns. In the design method evaluation, ZR04 and ZR08 are overall excellent, with ZR04 indicator ①, ZR08 indicator ①②, and two indicators ③ all receiving 7 points. The similarity between the generated pattern and the target style color distribution is 0.85, and the similarity between the texture feature vector is 0.82.
