StarGAN is a novel generative adversarial network (GAN) designed for multi-domain image-to-image translation. Unlike existing methods that require separate models for each pair of domains, StarGAN uses a single model to handle multiple domains, making it more scalable and efficient. The key innovation is the introduction of a mask vector, which allows the model to focus on specific domain labels during training, enabling it to learn from partially labeled datasets. StarGAN's architecture includes a generator and a discriminator, where the generator takes both the input image and the target domain label as inputs, and the discriminator classifies the generated images into their corresponding domains. The model is trained using adversarial loss, domain classification loss, and reconstruction loss to ensure high-quality and realistic translations. Experimental results on the CelebA and RaFD datasets demonstrate StarGAN's superior performance in facial attribute transfer and facial expression synthesis tasks, outperforming baseline methods in both qualitative and quantitative evaluations.StarGAN is a novel generative adversarial network (GAN) designed for multi-domain image-to-image translation. Unlike existing methods that require separate models for each pair of domains, StarGAN uses a single model to handle multiple domains, making it more scalable and efficient. The key innovation is the introduction of a mask vector, which allows the model to focus on specific domain labels during training, enabling it to learn from partially labeled datasets. StarGAN's architecture includes a generator and a discriminator, where the generator takes both the input image and the target domain label as inputs, and the discriminator classifies the generated images into their corresponding domains. The model is trained using adversarial loss, domain classification loss, and reconstruction loss to ensure high-quality and realistic translations. Experimental results on the CelebA and RaFD datasets demonstrate StarGAN's superior performance in facial attribute transfer and facial expression synthesis tasks, outperforming baseline methods in both qualitative and quantitative evaluations.