In this paper, a progressive creative image intelligent generation method combined with contrast learning is proposed, based on progressive growth generative adversarial network, using progressive growth training method to generate high resolution creative images. The contrast learning loss function based on the target key corner point features is constructed, and the multi-scale image feature loss function is combined to improve the quality of image generation. Based on the multi-dimensional validation of CIFAR-10 and CelebA datasets, explore the applicability of this paper’s method in visual communication design. The improved PGGAN algorithm Inception Score reaches 8.91±0.04, significantly outperforming all compared algorithms. When the optimal control parameter γ = 0.5 is adopted, the improved PGGAN algorithm performs better than the algorithms that introduce the style loss and the contrast learning loss alone. The FID minimum is reached about 40 steps earlier than the original model, and the generation quality and convergence speed are significantly improved. The subjective evaluation shows that this paper’s method tops all three scores, with a composite score of 4.51±0.03, and the image generation results are more creative and expressive. The method in this paper has excellent performance in image generation and can provide new possibilities and directions for visual communication design.
