This repository implements GANs N’ Roses, a generative adversarial network for image-to-image translation. The model separates content codes from face images and randomly sampled style codes to produce diverse anime images. It derives an adversarial loss to ensure both diversity and correctness in the generated outputs. The approach extends to video-to-video translation without requiring video training data, learned purely from still images.