We describe here a method to build fast and realistic synthetic images, which does not require a tedious a priori modeling of the scene to be displayed. We only use photos or videos of a real scene taken from some viewpoints, which in some way we 'wrap' to get new (synthetic) images of the same scene. This way, we can virtually see and move inside the 3d scene. The process consists of two steps: we first need to registrate the so-called reference views, using computer vision techniques. Then we just project the obtained model onto the plane of a virtual camera; this last step is fast and gives realistic results, allowing to use the method for instance to simulate the environment of a tele-operated device. 1 Introduction Synthetic images have proven useful for telemanipulation: synthesizing the environment of the operated device is needed for visualization, simulation, and better remote positioning.