In virtual reality systems, it is important to get both reality of the 3D scene and high interactivity of the system simultaneously. Although the Levels of Detail (LOD) approach is a popular solution, it requires great efforts to define multiple polygon models for each object. We developed a simple technique to get such interactivity based on the scene graph, which is used in VRML and Inventor as a scene database. We applied a fractal-based information filtering technique called Fractal Views to the scene graph. When users are not moving, the system renders all objects in the scene graph. When users start moving, the system renders a current focused object and its neighbors (in the scene graph).