Design and application of art creation education system based on generative adversarial network

In response to the limitations of traditional art education, this study proposes a generative adversarial network-based system for supporting artistic creation and teaching. A novel Hierarchical Attention GAN (HAGAN) model was designed and evaluated on a curated dataset of 5000 artworks across various styles. Quantitative metrics (SSIM 0.8325, PSNR 31) and human feedback from 60 participants confirmed HAGAN’s superior quality and diversity over DCGAN and Variational Autoencoder (VAE). Students rated generated works highly in creative inspiration (avg. 4.125), while teachers affirmed their value in instructional support (avg. 4.225). Surveys and Likert-scale protocols ensured reliable subjective evaluation. Although promising, real-world deployment challenges such as device compatibility, computational load, and network access are discussed as key limitations. Future work will address scalability and integration into diverse classroom environments.