One could aptly say that Nils Thuerey’s experiences in computer modeling and simulation lean toward the dramatic: He and three colleagues won an Oscar for Technical Achievement in 2012 from the Academy of Motion Picture Arts and Sciences for developing an algorithm to create fast and controllable smoke simulations and explosions on film, and a 90-second scene from a feature film involving a burning horse and lots of slow-motion fire stands out in his mind as his most challenging visualization of late.
The horse and fire sequence was from “Rise of an Empire,” the prequel to the movie “300” about the Battle of Thermopylae in 480 B.C. And specifically, Thuerey and colleagues Theodore Kim, Markus Gross, and Doug James won the Oscar for “the invention, publication and dissemination of Wavelet Turbulence Software.” The application employs a technique that has “allowed for fast, art-directable creation of highly detailed gas simulation, making it easier for the artist to control the appearance of these effects in the final image,” the description of the award reads. The software has been used in about 30 feature films, including “Avatar” and “Iron Man 3.”
“It’s great to see technology making a real impact in industry — to see that it’s useful, and ultimately being able to watch it on the big screen,” Thuerey said.
In a recent talk in San Diego at the XSEDE13 conference — the annual meeting of researchers, staff and industry who use and support the U.S. cyberinfrastructure — Thuerey provided an overview of the technical methodology involved in special-effects turbulence modeling and simulation research.
Simulating and Iterating
The development of simulated special effects is accelerated by high-performance computing (HPC), with speed advantages afforded by the parallelization of data and the use of graphics processing units (GPUs). Thuerey explained that in movie-making, right next to art direction (project control) in terms of importance is the time required to run a special-effects simulation, and the average turnaround of one-half day allowed by HPC seems to suit film artists.
The control aspect of film production also includes rendering the chaotic turbulence in fire, smoke and water and its complexity with as much realism as possible. The general approach to doing that, Thuerey said, is to start with a coarse and fast simulation and turn it into one that is detailed and of high resolution.
Fluid simulations — which in this context can refer to fire, smoke or water — serve as the base layer of a special effect, to which overlays, textures and particles are added. “You can have each layer approved, and then the simulation can remain ‘locked’ and unchanged,” Thuerey said. “In general, all movies iterate a lot: an artist produces different versions with feedback from supervisors and clients until everyone is happy — or as happy as possible.”
To add the details to effects, the researchers examine what are referred to as octaves in wavelet (small wave) turbulence. Metaphorically akin to musical octaves, these separate the different sizes of vortices (whirling masses) in a turbulent flow, and the large ones can be broken down into smaller and smaller ones. “We need vortices of very specific sizes that we can correctly blend in with those of the coarse simulation,” Thuerey explained.
The workflow for special-effects creation devised by Kim, Thuerey, James and Gross consists of the iterative steps of conceptualizing the artistic goal and developing the coarse simulation, followed by the execution of the one-time actions of detail detection, the tracking of motion and the application of turbulence.
More Particles, More Realism
Thuerey and colleagues Tobias Pfaff of the computer graphics laboratory ETH Zurich and Jonathan Cohen and Sarah Tariq of NVIDIA developed a scalable method of resolving the fine details of turbulent flows and published a paper entitled “Scalable Fluid Simulation using Turbulence Particles” for SIGGRAPH Asia 2010. Thuerey discussed highlights from the paper during his talk at XSEDE13, relating how in their methodology they use what’s called a two-equation K-epsilon model to compute the transport of turbulent energy, which they integrate into a base flow of smoke. In the next step, the researchers add particles for greater realism, without changing the overall flow of the effect. The end result is turbulent, billowing smoke. Thuerey added that the faster speed afforded by GPUs makes the computing of more interesting flows possible.
“The ‘classical’ use of turbulence models in computational fluid dynamics is to gain knowledge about, say, averaged quantities, for example,” Thuerey explained. “For graphics, we are more interested in synthesizing the turbulent flow over time to generate images with it. So there’s quite a difference in the goals for each direction.”
“Anything is Possible”
HPC can help not only with an explosion taking place in the foreground of the screen but also in the background in the form of what are called “invisible VFX.” “The easier it is to create these effects, the more we can use them in all parts of a scene,” Thuerey said. One example he gave during his talk was the computerized addition of bruises on an actor.
With the computational power and the advances made possible by research in special effects, “anything is possible, but it can be pretty expensive,” Thuerey said.
“The effects require very heavy computations, and the outcome is difficult to predict,” he explained. “So it takes many iterations to reach the desired shape, motion, etc.”
The content of Thuerey’s research and talk corresponds nicely with much of the activities taking place across the XSEDE ecosystem, according to XSEDE13 Technical Program Chair Amit Majumdar. “Scientfic visualization of terabytes to petabytes of data, produced by HPC simulations, is a big part of the end-to-end science process for XSEDE users,” he said. “Nils’ talk was excellent, as he discussed how he combines advanced algorithms and knowledge of domain science, such as turbulence, to generate these amazing visualizations.”
Thuerey praised the merit of XSEDE, saying, “It’s great to have such a strong organization for the high-performance-computing field.”
The annual XSEDE conference, organized by the National Science Foundation’s Extreme Science and Engineering Discovery Environment with the support of corporate and non-profit sponsors, brings together the extended community of individuals interested in advancing research cyberinfrastructure and integrated digital services for the benefit of science and society. XSEDE13 was held July 22–25 in San Diego; XSEDE14 will be held July 13–18 in Atlanta. For more information, visit https://conferences.xsede.org/xsede14.