This chapter systematically introduces image translation tasks, covering fundamental concepts, classifications, and core GAN-based methodologies. Image translation is defined as transforming images between domains (e.g., RGB to anime). Tasks are categorized into global (e.g., stylization, segmentation) and local (e.g., facial attribute editing), as well as supervised (paired data) and unsupervised (unpaired data) paradigms. Key supervised models include Pix2Pix (using U-Net generators and PatchGAN discriminators with L1 reconstruction loss), Pix2PixHD (enhancing resolution via multi-scale generators/discriminators), and Vid2Vid (video translation with optical flow constraints). Unsupervised approaches focus on domain alignment (e.g., CycleGAN with cyclic consistency loss) and latent space sharing (e.g., UNIT combining VAE and GAN). A practice on image coloring is demonstrated for deep understanding the details of Pix2Pix framework.