Performance Analysis of Digital Creative Visualization of Visual Images based on Artificial Intelligence Algorithms

With the rapid development of the digital creative industry, there has been a dramatic surge in the demand for processing massive amounts of data and generating personalized visual content. Traditional visual image generation technologies struggle to meet these demands, particularly in terms of detail representation, semantic consistency, and the efficiency of image generation. In this study, we propose a method based on a Dynamic Attention-enhanced Generative Adversarial Network (DA-GAN). By constructing a network architecture that includes a progressive upsampling generator embedded with a CBAM module and a discriminator combining PatchGAN with global average pooling, we optimize the model using an improved Wasserstein GAN adversarial loss, a perceptual loss based on a pre-trained VGG-19 model, and a style loss calculated via the Gram matrix. Experimental results demonstrate that compared to methods such as StyleGAN2 and SAGAN, DA-GAN achieves significant improvements in metrics like PSNR and SSIM, exhibits a lower FID value, and generates images with richer details and stronger semantic consistency. Additionally, DA-GAN offers notable advantages in computational efficiency and exhibits lower growth in memory consumption. This study provides an effective solution to the challenges faced by traditional visual image generation technologies in detail representation, semantic consistency, and generation efficiency, with broad application prospects.