Abstract Thestyle-basedGANarchitecture(StyleGAN)yields state-of-the-artresultsindata-drivenunconditionalgenerativeimagemodeling.Weexposeandanalyzeseveralofits characteristicartifacts,andproposechangesinbothmodel architectureandtrainingmethodstoaddressthem.Inpart