Computer vision, as a technical means of “seeing” the world, is applied to modern art design, which can expand the audience’s perception of external reality and re-understand the process of art design. This paper examines the process evolution and creation method of generative art design by analyzing the characteristics of computer vision technology and visual communication design. Based on a capsule network and a generative adversarial network, the C-CapsGAN model for art design image generation has been established. Optimization functions for consistency, style loss, and smoothness loss have been designed. Quantitative analysis and survey research were used to analyze the effectiveness of the C-CapsGAN model in art design image generation. It was found that after the model convergence, the FID value was reduced by 20.36%, and the IS value was improved by 0.482 compared to CCME-GAN on the modern art design image dataset. The mean SSIM value of the six art design images was 0.611, and the subject’s favorite score for the model-generated art design images could be up to 3.715 points. By integrating computer vision technology with modern art design, the diversification of art images can be promoted, and the visual stimulation of art design can be enhanced for the audience.