Generative　Adversarial　Networks　(GANs)　have　rapidly　risen　to　prominence　in　the　sphere　of　deep　learning.　This　is　especially　true　when　it　comes　to　image　generation,　where　GANs　have　displayed　impressive　capabilities.　Over　time,　as　researchers　have　grappled　with　the　challenges　posed　by　the　original　GAN　model,　a　plethora　of　GAN　variants　have　been　introduced.　These　are　tailored　to　counteract　training　instability,　mode　collapse,　and　various　other　challenges　intrinsic　to　the　base　GAN　model.　Despite　the　extensive　variety　of　GAN　techniques　available,　there　is　still　a　knowledge　gap　when　it　comes　to　understanding　their　relative　performances　on　specific　datasets.　For　many　practitioners,　choosing　the　right　GAN　model　for　a　given　dataset　remains　a　trial-and-error　endeavor.　In　an　attempt　to　shed　light　on　this　matter,　our　research　undertakes　a　deep　dive　into　three　of　the　most　notable　GAN　techniques:　the　classic　GAN,　the　one　built　upon　Wasserstein　Distance,　often　abbreviated　as　W-GAN,　and　its　subsequent　evolution,　WGAN-GP.　©　2023　IEEE.