Artificial intelligence-generated content (AICG) is increasingly being applied in the cultural and creative industries. This paper focuses on the application of AICG in enhancing the diversity of generated content and reducing the risk of plagiarism. To control image generation attributes, a generative adversarial network (StyleGAN) with a style transfer module was designed to improve the training effectiveness of the model. Modules such as improved modulation and demodulation, and a jump-based structure were incorporated to construct the StyleGAN2 network, optimizing the detail of generated images. The K-means algorithm is used to perform optimal style clustering of the generated images. Research shows that when the truncation point number of the StyleGAN2 network is set to 10, the FID value ranges from 4.312 to 4.653. The highest score for generated images reaches 9.027, with a modification rate not exceeding 1%. The generated images are clustered into five categories, with contour coefficients ranging from [0.0, 1.0].
