Using Pix2Pix conditional generative adversarial networks to generate personalized poster content: Style transfer and detail enhancement

In the process of generating personalized poster content, there are often problems such as inconsistency, blurring, and distortion between the generated image and the original image content, exaggerated image styles that do not meet the personalized needs of users. In response to these issues, this article aims to use Pix2Pix Conditional Generative Adversarial Network (Pix2Pix-CGAN) to solve problems such as image blur, distortion, and content inconsistency in personalized poster generation, in order to improve the quality of generated images and meet users’ personalized needs. This article combined L1 loss and adversarial loss optimization models, proposed a style transfer mechanism for personalized poster generation models based on Pix2Pix, and proposed corresponding detail enhancement strategies. The U-Net architecture was used to improve the generator structure, and PatchGAN was used to ensure local detail clarity and image quality. The experimental results showed that the color consistency (CC) of the personalized poster generation model based on Pix2Pix was 0.97, the Style Preservation Index (SPI) was 0.93, the PSNR was 35 dB, and the SSIM was 0.92. The Detail Enhancement Factor (DEF) was 0.88, and the high-frequency component recovery (HFCR) was 0.09, which was at an excellent level compared to other style transfer methods and detail enhancement strategies. The results of this study indicate that the personalized poster generation model based on Pix2Pix exhibits significant advantages in style transfer and detail enhancement, effectively improving the quality of generated images and meeting users' personalized needs.