Automatic Generation and Optimization of Visual Design Elements Based on Generative Adversarial Networks

With the rapid development of artificial intelligence technology, generative adversarial networks (GANs) have shown remarkable potential in the field of image generation and optimization. However, existing methods often limit themselves to single scenarios or lack a comprehensive consideration of diversity and aesthetic optimization when dealing with visual design elements. To address this issue, this paper proposes an innovative algorithm framework based on GANs (AGOD-GAN) for the automatic generation and optimization of visual design elements. The framework introduces a multimodal feature extraction module, achieving multidimensional feature modeling and prominence enhancement of visual design elements by combining self-attention mechanisms and style transfer techniques. In the generation phase, an improved generator structure is adopted, introducing a hierarchical feature fusion strategy to enhance the quality and diversity of generated elements. Additionally, by introducing an adaptive discriminator, the weights of the discriminative network are dynamically adjusted, significantly enhancing the model's adaptability in complex visual tasks. Furthermore, to address aesthetic bias in the generated results, this paper designs an iterative optimization mechanism based on user feedback, which can further optimize the artistic and functional aspects of the generated elements while maintaining style consistency. Experimental results show that the proposed method exhibits excellent performance in visual design tasks, achieving significant improvements in both generation quality and diversity compared to existing mainstream methods, and also demonstrating strong practicality in aesthetic optimization. This study provides a new technical approach for the automation of visual design, with broad application potential.