The pixel2style2pixel (pSp) framework uses a novel encoder network to directly generate style vectors fed into a pretrained StyleGAN generator. Unlike traditional invert-then-edit approaches, this encoder handles translation tasks by encoding input domains directly into the latent domain. The method requires no adversarial training and supports multi-modal synthesis through style resampling, enabling various image translation tasks including frontal face synthesis, super-resolution, and sketch-to-image translation.