Self-Distilled StyleGAN: Towards Generation from Internet Photos Ron Mokady Michal Yarom Omer Tov Oran Lang Daniel Cohen-Or Tali Dekel Michal Irani Inbar Mosseri Abstract StyleGAN is known to produce high-fidelity images, while also offering unprecedented semantic editing. However, these fascinating abilities have been demonstrated only on a limited set of datasets, which are usually structurally