In order to address problems with image denoising, numerous approaches have been developed. Generative models arose as one of the most extensively researched areas in machine learning, particularly deep learning. Generative Adversarial Networks, a recent major discovery in the generative subject matter, are very fascinating. This research presents a novel approach for image denoising, notably for single noisy images, based on generative adversarial networks (GAN). The observations and statistics on the corruption process serve as the only source of information. The procedure optimizes the GAN as the prior and seeks to determine the maximum a posteriori (MAP) estimate. For creating realistic reconstructions from a corrupted observation, the generator adds a penalty term to impose the reconstructions on the associated observation. We demonstrate our methodology using CelebA dataset with various sizes and degrees of degradation. The suggested approach provides an alternative to unsupervised denoising and delivers outcomes that are on line with the state-of-the-art in noise reduction for generative models.