The GPT-4 version was officially released in 2023, and generative models are also developing rapidly with the development of artificial intelligence technology. However, with the extensive use of generative models, people have found that the output of them are unreliable. The design capabilities of these generative models need to be further explored. This study aims to evaluate the design capabilities of three generative models in the field of illustration design. Employing a questionnaire survey methodology, we analyzed the collected data using the Wilcoxon signed-rank test. Our focus was to assess three generative models in illustration design, specifically examining their capabilities in theme matching, color matching, artistic style, detail performance, creativity, and overall impression, and comparing these to the levels achieved by human designers. The results indicated that these generative models reached the benchmark set by human groups in all six aspects of illustration design. Notably, DALL·E 3 has advantages over artificial intelligence in terms of artistic style, detail representation, creativity, and overall impression. This research guides designers in using these types of generative models and also provides a direction for developers to continue to improve the design capabilities of generative models.