An era of creativity and invention has recently been enabled by the advent of AI-generated visuals but there is scope of research to increase the realism of these images. In this research, the proposed method aimed to enhancing the realism of AI-generated images using feature extraction and GAN-based enhancement techniques. This study begins with AI-generated images produced by Copilot, a cutting-edge large language model, which serves as the initial input and applies the models, i.e. VGG16 and ResNet50 to extract high-level features from the images. GAN model using the generated images from the models which is further fine-tuned with the hyper parameters for the specific dataset, to enhance the realism of the images through advanced techniques such as texture synthesis, color correction and structural adjustments, and various preprocessing techniques. The upgraded images underwent evaluation using metrics, i.e. Histogram Similarity, PSNR (Peak Signal-to-Noise Ratio), and SSIM (Structural Similarity Index), which revealed significant improvements in image quality and realism. The results show a significant gain in precision as well as detail when utilizing GANs to transform AI-generated images into realistic, human-like images. This breakthrough technology has useful applications in virtual reality, gaming, and digital media.