Image inpainting is an essential technique within the field of computer vision aimed at restoring missing or damaged areas of an image. This process has a wide range of applications, including image restoration, removal of unwanted objects, and enhancement of visual appeal. Conventional methods, such as diffusion-based and exemplar-based techniques, often fall short in their ability to create realistic textures and complex patterns, particularly when substantial portions of the image are absent. The emergence of deep learning, particularly Generative Adversarial Networks (GANs), has shown impressive capabilities in generating visually consistent and contextually appropriate content for these missing sections. This survey provides a thorough examination of the latest developments in image inpainting utilizing deep learning methodologies, with a specific emphasis on GAN-based frameworks. We analyze different architectures, discussing their advantages and drawbacks, as well as the evaluation metrics employed to assess the quality of inpainting. Furthermore, the paper outlines the challenges encountered by current models and proposes possible avenues for future research aimed at achieving greater realism and resilience in image reconstruction endeavors.