In recent years, the science of computer vision has seen a significant increase in researchers interested in studying facial hallucinations. This endeavour continues to be one of the most difficult to do because of the intricate and delicate structure of the human face. In this paper, an End-to-end Generative Adversarial Network for Face Hallucination based on extending ESRGAN method is proposed. The propsed model uses Residual in Residual Dense Block (RRDBNet)  in its generator to improve performance compared to ESRGAN benchmark model. The Generator creates visually appealing hallucinated face pictures. At the same time, the enhanced discriminator assesses the degree of realism in the input images. During the training phase, the mbileNet model can extract features. The suggested model will be able to conceive face representations that are perceptually credible because it will be able to have an advantage over ongoing competitive learning. Extensive studies conducted on huge face datasets demonstrate that the introduced method performs noticeably better than existing methods considered to be state-of-the-art.