Computer Graphics World

n n n n CGI Creating Cloud World "From the effects point of view, our biggest challenge was the cloud world," says Ken Bielenberg, visual effects supervisor. "In the past, we typically used matte paintings for clouds unless we wanted more dynamic stereo 3D clouds. But the systems we developed for 3D clouds were difficult to use, the data was too large, and they were unwieldy. So, we ended up doing only a handful of shots. On this film, we did around 40 shots with volumetric clouds, and they worked out really well. Brett Miller, our effects lead, was instrumental in making that possible." Miller spent a year and a half leading that effort, which relied almost entirely on in-house tools: a simulator called Flux, volume modeling tools that sit inside Side Effects Software's Houdini as a plug-in system, a new sparse volume format, and volumetric shaders. Puss, Humpty, and Kitty run around on top of clouds shaped like hills in a pas- toral English countryside or maybe the South of France. In the distance, some clouds have shapes like trees. "We see fluffy stuff swirling around, as if they were on a stage with dry ice," Miller says. "The clouds close to where they stand are packed and sculptural; they react to their movement. Initially, we wanted to cre- ate clouds with a photoreal look, but because the clouds represent childhood dreams, the color palette shifted from whites and blues to pinks, oranges, and gold. They look a little stylized and cartoonish." Modelers created the cloud shapes, and then the effects artists converted the shapes into particle-level sets. "Once we had the volumes modeled, we could take pieces and run them through Flux, our gas simulator," Miller says. "By sim'ing part of a cloud, we could give the impression that it's moving, without changing the look." Flux, which sits inside Houdini, works solely with voxel grids; voxel grids are the only inputs. "So, it's really flexible," Miller says. "If you want a new force field, you create a voxel grid. You don't have to write a plug-in to do a field. Emitters are voxel grids. Fields, collisions, everything." Everything, including characters. "We knew we couldn't take a character and replace it with a hierarchy of articulated rigid bodies to do the collisions," Miller says. "We really wanted deformations to push the clouds around. So we converted our character meshes into level sets and used those level sets to mask off velocities in the gas simulator, which gave the impression of collisions. Ron Henderson [R&D manager] had the idea of using voxelized data as the collision object. Rather than using a thin shell mesh, we were using the entire interior of the character." To add complexity in the motion, the artists sometimes added small collision blobs only to cause movement in the fluid; they didn't render the blobs. "It was a steep learning curve for people coming from [Autodesk] Maya, where they dropped tools into a voxel grid," Miller says. "Flux makes complex things simple, but it takes the same amount of work to make simple things. We're still figuring out the best way to use it to do simple things." Using a sparse volume format that R&D effects engineer Ken Museph wrote, the volumes could be huge. "One shot had a volumetric grid that was 900 by 500 by 15,000," Miller says. "But the format managed it really well. We couldn't have done this without it." Also helping to manage the huge area were new methods for speeding ren- dering with the sparse format. "We do a lot of work with frustum buffers," Miller says. "With these big cloudscapes, we have a continuous level of detail, and some were miles deep." To solve a typical problem with frustum buffers—that is, shadow issues created when panning the camera—global effects developer Devon Penny wrote a sys- tem in which it was possible to rasterize a simple orthographic buffer in addition to the frustum buffer. "The coarse buffer was never visible in the camera frustum," Miller says. "But it handled all the shadows outside the frustum buffer. It was very helpful." –Barbara Robertson 14 October/November 2011 Character TDs needed to rig Humpty's face so that, on one hand, it didn't look like the egg would crack when he smiled, and on the other, it would look rubbery. than before, and layers of dynamic perfor- mance, so the fur responds better to his move- ment," Loofbourrow says. "He has little tufts of hair on his elbows and in his ears, neck fur, and other little stuff that makes a richer look." The character TDs added controls to the rig for areas that the animators needed to see move or wanted to move themselves—the whiskers, for example; otherwise, the charac- ter effects department handled the dynamics. "We used some of our traditional fur smooshing techniques that we developed way back for Shrek," says Bielenberg. "But for this film, the effects department developed a [Side Effects] Houdini-based pipeline to do the hero fur smooshing. The great part about that was they could do four or five different types of sims together." One simulation might cause the fur to smoosh and avoid Puss's belt or boots. An- other might add wind. A third would move select parts of the fur a little more than others. A fourth would wet the fur. And so forth. "Before, we had to do the simulations seri- ally," Bielenberg says, "one, and then the next. With Houdini, we could do all the sims to- gether. Also, we could view all the guide curves in Houdini, so the effects artists could see the volume of curves and push any we needed to fix by using deformations to move them into position." Before, they deformed the curves by using texture maps and fields, which meant they worked blind; the artists couldn't see the result while they worked. "We still have to spend time doing technical steps, but we're definitely getting to the point where we can focus more strongly on the cre- ative iterations," Bielenberg says. The reason is that hardware is faster, and software is better. The next-generation tools also allowed more complex models.

• Cover