Generative Adversarial Networks (GANs) enable generating photo-realistic images more successfully compared to other generative models. However, when the resolutions of the generated images increase, the stability and the diversity problems that usually occur in GANs, cause important problems in generating images with high quality and variety. In this study, we empirically examined the state-of-the-art cost functions, regularization techniques and network architectures that have recently been proposed to deal with these problems, using CelebA dataset. In order to compare the numerical performances of the models, we used Frechet Inception Distance (FID) metric, which performs well in comparisons with the images in terms of blur, noise, distortion and diversity. As a result of improvements that are made based on the reference model, the FID score is reduced from 137 to 9.4.