Creative generation and evaluation system of art design based on artificial intelligence

Abstract

This study examines the performance of various AI models in artistic image generation, including generative adversarial networks (GANs), convolutional neural networks (CNNs), hybrid models, and style transfer techniques. The results show that hybrid models excel in quality, diversity, and artistic expression while maintaining efficient generation times. This suggests that hybrid models combine the generative creativity of GANs with the structural precision of CNNs, making them suitable for applications requiring innovation and visual fidelity. GANs demonstrate exceptional creativity but fall short in artistry and output diversity, making them ideal for exploratory and experimental art creation. CNNs offer stability and detail accuracy but lack innovation, making them useful for tasks requiring consistency, such as style preservation. Style transfer methods preserve artistic characteristics but exhibit lower diversity and creativity, limiting their application to predefined stylistic transformations. The study contributes to the ongoing discourse on improving AI's capacity for creative expression while balancing creativity, diversity, and computational efficiency.