In this article we present a visualisation algorithm based on image warping. A source data for the algorithm is a set of images with colour and depth information rendered for the observer located in a number of reference positions. For each position there are 6 images rendered for the directions of axes forming the "image cube". The algorithm computes a 3D Delaunay triangulation on these positions and divides the space inside the convex hull into tetrahedra. During the visualisation process the observer situated in one tetrahedron sees warped images from the vertices of this tetrahedron. We also describe the automatic algorithm for selection of new reference positions to improve the quality and remove "holes". For both algorithms we utilize 3D hardware acceleration. Keywords image-based rendering, warping, visualisation, Delaunay triangulation