In the metaverse environment, avatars perform various roles as alternative representations of individuals. Avatars can be expressed from the perspectives of enhancing immersion and providing flexible security measures within the metaverse space. This study aims to adjust the representation of avatar details based on the avatar"s role in the virtual space. By tuning the hyper-parameters(such as Random Seed, Style Embedding Vector, Batch Size, Number and size of layers in the generator network, Number and size of layers in the discriminator network) of the styleGAN algorithm, a system is designed where the level of immersion, privacy preservation, and data exchange on the network can be controlled to automatically adapt the level of detail in the avatar.