Fig. 1. Qualitative results of HiFaceGAN, SRGAN, Pix2Pix, AxialGAN and the proposed ANYRES on the ARL-VTF dataset. We decrease resolution in each row (re-scaled to 128×128). While previous methods are impaired to super resolve facial images for a given resolution by using one specific network for each resolution, our proposed ANYRES achieves a balance between realism and fidelity across resolutions with solely one unified network.