A method for delivering media from a server to a client device over a network is disclosed. An Virtual-Reality (VR) scene application running on the server generates a virtual VR 3D environment containing 3D models. The server checks the status of each 3D model in a predetermined order, and then, only those 3D models which are not pre-stored in the client device will all be rendered by the server into a left eye frame and a right eye frame of 2D video stream. The server then sends to the client device the frames and meta data of the 3D models which are pre-stored in the client device via the network. The client device uses a combined VR frame of these frames as a background for rendering the 3D models which are pre-stored in the client device so as to generate a mixed VR frame of video stream for output.