Abstarct: Nowadays, having high quality images and videos is undoubtedly of a matter of importance. Super-Resolution technique is a software approach within the scope of signal processing, employed for enhancing the quality of an image by increasing the number of pixels along with decreasing the noise. In this technique, a high-resolution image will be created from one or multiple low-resolution images.In this thesis, we will introduce a video super-resolution method founded on learning approach and baxsed on a convolutional neural network, in which a trainable Optical Flow network is used for gathering the information between the current frxame and the previous one. The extra information received from the Optical Flow network, will be added to the current low-resolution frxame. Afterward, this data will be sent to the Super-Resolution block where the current frxame is super-resolved by a deep neural network. finally, the reconstructed high-resolution frxame as a generator block will be given to a discriminator block in a Generative Adversarial Network (GAN) frxamework which will finally result in presenting a Photo-Realistic Super-Resolved frxame. The suggested approach, taken through using three trainable blocks, employing convolutional neural networks, receiving data from Optical Flow and applying Generative Adversarial Networks, has succeeded in magnifying the video 4 times without any conspicuous reduction in quality. In the end, the proposed model (FRGAN-VSR) was compared with the SRGAN model (Omitting Flow Network from the proposed method) and also FRVSR (Removing Generative Adversarial Network from our model) and evaluated in the Training, Evaluating and Testing phases. To assess the visual quality, we tested our model baxsed on PSNR and SSIM measurements. The proposed model has also been tested in a few sample frxames along with a few multi-frxame videos. Furthermore, we analyzed the temporal profile of the model and compared its temporal consistency with other models. The evaluations and comparison with previous methods demonstrate that the proposed model can outperform the current state of the art.