An image rendering device realizes stereoscopic viewing of composite images generated by compositing background three-dimensional models and foreground three-dimensional models each defined in three-dimensional modeling space. A texture mapping unit converts background image data into two or more viewpoint textures and maps each viewpoint texture to a background three-dimensional model in the three-dimensional modeling space. A viewport conversion unit extracts, for each of the two or more viewpoint textures, a viewport image from the background three-dimensional model mapped with the viewpoint texture and from the foreground three-dimensional model. By comparing a parallax of the foreground three-dimensional model occurring in stereoscopic playback with a parallax of the intra-texture object occurring in stereoscopic playback, the device determines depth relationship between the foreground three-dimensional model and the intra-texture object and adjusts positional relationship between the foreground three-dimensional model and the intra-texture object based on a result of the determination.