A deep learning framework called generative adversarial networks (GANs) employs - a discriminator and a generator- that compete with one another to produce realistic yet previously unobserved examples. They have gained popularity as a study area recently, especially for image processing and synthesis, which has resulted in numerous advancements and applications in many other fields. Surveys on GANs are important, especially for individuals who want to start on this issue, given the abundance of published works and attention from professionals in various fields. With an emphasis on its use in the creation of images, we discuss the fundamentals and significant architectures of GANs in this paper. We also highlight various challenges that have been encountered when developing GANs designs, including mode reach, stability, convergence and metric-based image quality assessment.