The state-of-the-art methods in accelerating dynamic Magnetic Resonance (dMR) Imaging rely on sparse and/or low-rank priors. We propose a novel manifold driven framework that exploits the manifold smoothness priors to highly accelerate data acquisition in dMR. We postulate that images in dMR lie on or close to a smooth manifold and learn the manifold geometry from the navigator signals. Capitalizing on the learned manifold, we develop two regularization loss functions and subsequently build a framework to reconstruct dMR images from highly undersampled k-space data. The proposed method is shown to be superior than competitive methods in different data sets.