Generative models have emerged as a cornerstone of machine learning, offering powerful tools for data generation, image transformation, and feature extraction. This paper presents an in-depth exploration of three prominent generative models: Generative Adversarial Networks (GANs), Neural Style Transfer (NST), and Autoencoders (AEs). GANs, with their unique architecture of a Generator and Discriminator, have evolved with variations like DCGAN, CGAN, WGAN, and CycleGAN, each enhancing performance in various tasks. Neural Style Transfer, rooted in deep learning, enables the blending of content and style across images using models like VGG-16 and VGG-19, and its advanced variations like Fast Style Transfer and Multiple Style Transfer allow for real-time applications in creative industries. Finally, Autoencoders provide an unsupervised framework for dimensionality reduction and feature learning, with specialized versions like Variational Autoencoders (VAEs) for anomaly detection, Contractive Autoencoders (CAEs) for feature extraction, and Adversarial Autoencoders (AAEs) for data generation. This paper reviews the architectures, key features, and applications of these models, demonstrating their relevance and versatility.