In this study, generative adversarial network is used as the basic architecture, and the multi-head attention mechanism is introduced to enhance the model’s ability to perceive and process image features. The image generation process is optimized by bilinear interpolation to further enhance the detail expression of character design. The generation efficiency of the model and the quality of the IP image are improved by the improved network structure. A personalized recommendation model with implicit feedback and explicit feedback is also used to achieve targeted placement of IP image characters for agricultural and sideline products cartoons. The study combines the local characteristics of Jilin Province, taking Jilin rice as an example, and designs two rice brand IP images with regional characteristics, “Rice Xiaoji and Rice Xiaoling”, which have a good migration effect. When the recommended list length is Top=10 and 20, the recommendation effect of internal diversity of Jilin rice brand reaches 83.47% and 89.09% respectively, and the recommendation effect of overall diversity reaches 88.43% and 95.31% respectively. It can be seen that the method of this paper can improve the market competitiveness of agricultural and sideline product brands in Jilin Province, which provides a technical path and practical reference for rural revitalization in Jilin Province.