This chapter comprehensively explores the advancements and methodologies in image generation using Generative Adversarial Network. It begins by outlining key applications of GANs, including training data expansion, synthetic data refinement, and creative content generation, highlighting their role in reducing manual data collection efforts and enhancing model generalization. The chapter then delves into foundational frameworks such as Deep Convolutional GAN (DCGAN), detailing its architecture, transposed convolution mechanisms, and limitations in generating low-resolution images. Subsequent sections focus on conditional GAN variants (CGAN, InfoGAN, ACGAN), which integrate explicit or implicit control over generated outputs through labeled or latent variables. Multi-scale GAN architectures, including LAPGAN and Progressive GAN, are introduced to address high-resolution image synthesis challenges via hierarchical residual learning. StyleGAN, a landmark model for attribute disentanglement and high-fidelity face generation, is analyzed in depth, emphasizing its mapping network, adaptive instance normalization (AdaIN), and style-mixing techniques. The chapter also covers advanced topics such as multi-discriminator and multi-generator frameworks to mitigate mode collapse, as well as GAN applications in data augmentation (BAGAN) and simulation refinement (SimGAN). Finally, practical implementations of DCGAN and StyleGAN are demonstrated, including code interpretation and training details.