The Generative Adversarial Networks have been emerging as one of the most powerful tools for image synthesis, also they often face critical training issues and instabilities such as mode collapse and vanishing gradients. These issues lead to convergence and reduce diversity in generated outputs. This work deals with StableGAN, a hybrid approach that integrates Wasserstein GAN with Gradient Penalty (WGAN-GP), improved gradient Penalty Computation, architectural improvements and learning rate scheduling to achieve more suitable training dynamics. Optimisation is done with adaptive Adam configurations, and regularization techniques are used to maintain Lipschitz continuity without excessive weight clipping. Experimental evaluation is performed on the CIFAR-10 dataset, for demonstration of improved quality of images, diversity and convergence speed compared to baseline WGAN-GP Models. Qualitative Matrices, visual samples and training loss curves validate the effectiveness of the proposed approach in mitigating mode collapse and vanishing gradients. This study also highlights the scalability of StableGAN for broad generative tasks while preserving computation efficiency.