In the context of the digital era, brand image design presents new characteristics and challenges. In this paper, the human visual system (HVS) and the visual perception characteristics of the human eye are first elaborated to provide a basis for multi-level visual communication design. Then, based on the generative adversarial network, PCGAN, a color migration model for brand graphics based on style coding, and SPGAN, a style transformation model for brand graphics, are constructed, and the performance of the models is verified and evaluated. The Frechette Distance (FID) value of SPGAN is 60.241, which is much smaller than that of the same type of scheme, the StarGAN model (120.467) and the StarGAN-v2 model (82.625), and the SPGAN model is significantly better than the other models in terms of training stability, indicating that it works best in the style migration task and generates images of optimal quality. In addition, according to the subjective questionnaire research results, appropriate color matching, reasonable visual layout and conspicuous brand logo are the key factors to enhance the attractiveness of the product and promote brand memory, so as to construct the digital brand image. Combining the two design models, this paper provides methodological references for the construction of digital brand images.