Designing StyleGAN2 Model for Generating 2D Pokemon Character

In this paper, we propose a 2D image generation model using StyleGAN2 applied in the creative design of publications for entertainment purposes. There are only 898 Pokemons created throughout 26 years. Therefore, we focus on generating new Pokemon to signicantly reduce their cost. We built a dataset of approximately 800 images on Kaggle, applied DiffGANAugment, and trained them using the styleGAN2 model. The Frchet inception distance (FID) score evaluates the generated images' quality. The results show that the proposed model achieves an FID score of 53.23 after 33 epochs that are applicable in real hardware.