The information accessing through magnetic resonance (MR) scanning for diversified diagnosis is increasing rapidly for the availability of information with multiple contrasts but the acquisition of some contrast the limit this. The noise and artifacts affect the image quality and the contrast is limited due to the scanning time and added noise during acquisition. Present techniques for multi-contrast synthesis use deterministic neural networks or nonlinear regression to learn a nonlinear intensity transformation between the source and destination images. The loss of high-spatial-frequency information in synthesized images can be a drawback for these techniques. Here, we suggest a novel method based on conditional generative adversarial networks for multi-contrast MRI synthesis. The suggested method provides improved synthesis performance through a pixel-wise loss for registered multi-contrast images and a cycle-consistency loss for unregistered images, while preserving high-frequency information via an adversarial loss. Utilizing data from nearby cross-sections, the synthesis quality is further enhanced. The suggested technique performs significantly better than earlier state-of-the-art techniques, as demonstrated by demonstrations on T1- and T2-weighted pictures of patients and healthy participants. With our synthesis approach, multi-contrast MRI tests can become more versatile and of higher quality without requiring longer exams.