Impacts of Generative AI on Style Diversity in Graphic Design Based on Deep Generative Adversarial Networks

With the rapid advancement of artificial intelligence (AI), Generative Adversarial Networks (GANs) have become a powerful tool in the field of graphic design. However, balancing the trade-off between style diversity and sample quality remains a significant challenge. This study proposes a novel framework for generating and evaluating graphic design style diversity using conditional GANs (cGANs). By incorporating regularization strategies and style embedding mechanisms, the framework allows precise control over both the style and the diversity of generated samples. Experimental results demonstrate that the proposed model effectively generates high-quality samples in both Retro and Minimalist styles. The Frechet Inception Distance (FID) scores of 22.5 for Retro and 28.7 for Minimalist, significantly lower than the baseline model's FID score of 34.2, indicate that the generated samples are visually closer to real data. Additionally, the Perceptual Path Length (PPL) score of 120.4 (compared to 178.7 for the baseline model) reflects smoother and more diverse distributions in the latent space. t-SNE visualizations further validate the broad coverage and clear separation of generated samples across different styles. This work provides a new paradigm for applying AI in the creative industry, with significant implications for the automation and diversification of graphic design.