Graph-based Fire Synthesis

We present a novel graph-based data-driven technique for cost-effective fire modeling. This technique allows com-posing long animation sequences using a small number of short simulations. While traditional techniques such as motion graphs and motion blending work well for character motion synthesis, they cannot be trivially applied to fluids to produce results with physically consistent properties which are crucial to the visual appearance of fluids. Motivated by the motion graph technique used in character animations, we introduce a new type of graph which can be applied to create various fire phenomena. Each graph node consists of a group of compact spatial-temporal flow pathlines instead of a set of volumetric state fields. Consequently, achieving smooth transitions between discontinuous graph nodes for modeling turbulent fires becomes feasible and computationally efficient. The synthesized particle flow results allow direct particle controls which is much more flexible than a full volumet-ric representation of the simulation output. The accompanying video shows the versatility and potential power of this new technique for synthesizing realtime complex fire at the quality comparable to production animations