This paper presents a server-based remote walkthrough system. The client is assumed to be a thin client, like a handset or a mobile device, with no strong processor but with some embedded video chip. The server holds the large environment, generates the frames, encodes and transmits them to the client. The encoded frames are transmitted as a video stream to the client, which then decodes the stream and displays it. We show how the computer generated frames can be efficiently encoded using layering techniques to yield a lighter stream, which enables its transmission over narrow bandwidth channels and minimizes the communication latency. To enable the interactivity of the system, the rendering engine generates the frames in real-time according to the client input, and feeds the frames to an accelerated video encoder based on the available optical flow.