A Survey on Generative Models for Image Synthesis: GANs, Diffusion Models, and Beyond

Abstract

Generative models have changed computer vision by enabling realistic and diverse image synthesis. This survey provides a comprehensive review of popular generative models; Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Diffusion Models. We examine their architectures, training methodologies, strengths, limitations, and applications. Key datasets and evaluation metrics are discussed, in addition to insights obtained from comparative analysis. Additionally, ethical considerations and research challenges such as training instability, mode collapse, and computational complexity are addressed and potential future directions are explored to guide research in efficient, interpretable, and safe generative modeling.