The goal of our research is teleconferencing with realistic sensation. Generating images with motion parallax is one of the key technologies which will make it possible to further human to human communication. To generate images with realism by CG, the number of vertexes of an object should be increased. However, to generate images with motion parallax, the images should be generated quickly. So, there is a tradeoff between generating highly realistic images and generating them quickly. In this paper, we evaluate the effect of delay time according to three parameters: response time of generating images for head movement and frequency of generating them, and amount of task in virtual space. We also present a method to generate CG images with realism at high speed. The method is to select from a database of different detailed models, one model based on the distance between the viewpoint and object, or between gazing point and object.