Computer Graphics World

November 2008 37 CG I n n n n too complex a structure for a single build- ing," explains Clarenc. "Instead, we created a library of parts, like rooftops, walls, and blocks, and assembled them by hand." e actual building destruction was done with a rigid-body dynamics system. "If you want to break a building into two pieces, you must also model the interior and break everything into small pieces," explains Emilien Dessons, special effects supervisor. "We tried to use an algorithm-based shat- tering technique, but the shattering was equal and ugly, so we looked for a chaotic shatter with a graphically appealing shape. We manually modeled everything, shatter- ing each piece with a reference drawing. We also added classically instanced particle systems to the destruction scenes to bring life to them, with a large amount of com- positing and different smoke techniques added in as well." Furthermore, the crew used two model- ing LODs, one with as few faces as pos- sible, and another with beveled angles so the team could automatically subdivide the objects depending on the needs during ren- dering. "You could probably say we had a third level, where the objects were detailed according to the textures for extreme close- ups," Clarenc says. Devil in the Details While atypical, the environments never- theless had to look realistic, in contrast to the character models, in keeping with the style of the movie. "Guillaume Ivernel's ref- erences were very detailed, and we had to translate that level of detail in every texture we created," says Fabien Polack, texturing supervisor. e environmental textures were fairly traditional, a combination of texture maps with simple shaders. "e challenge for us was more artistic than technical; we had to match Ivernel's vision and enhance it when- ever possible," says Polack. Due to time lim- itations, the artists created ground shaders that worked for almost every shot—from expansive shots to close-ups. ey also painted detailed textures but were careful to not fall into a common trap by "over-tex- turing" objects. "Like painters, we removed some material at the end of a texture to make it look more natural," he adds. Every set, Polack explains, was sur- rounded by a sphere mapped with a high- resolution matte-painted sky, provided by the matte department or Ivernel himself. Most of the time the reflections were real blurred reflections, and less often, environ- mental maps. Every map was created, at minimum, at 4k resolution, sometimes 8k, and on oc- casion, 16k. e group provided the tra- ditional set of textures—color, specular, bump, reflectivity, and, depending on the context, roughness, displacement (usually at 16 bits), and normal. en, each high- definition shader was converted to mid- (2k) and low-def (1k) depending on the shots. e bump factors were also driven by a "distance-to-camera" node. Despite using what Clarenc calls "a clas- sic modeling setup," the group still ran into some issues. "For instance, our renderer doesn't have an angle-threshold limit to cal- culate the phong smoothing, which means that all our low-def models were totally out of shape and all the details were completely lost. So we developed a tool to convert the Maya hard edges—which are nothing but an edge-bound angle limit for the phong smoothing—to 'real' hard edges that were vertex-based with doubled adjacent edges to stop the phong from averaging values across them. And, voila, we had our low- def Maya objects in their crisp, detailed glory. While it doesn't sound like a big deal, it soon became vital for the completion of the huge sets at the end of the movie." To render the images, Mac Guff had a 650-processor renderfarm available—small in comparison to Hollywood standards— though on average only 200 to 300 pro- cessors were used per shot. To keep the rendering times manageable, the crew was encouraged to keep each shot to a render- ing time of less than two hours per frame, no matter the number of layers. Neverthe- less, working in layers gave the artists more control over the final result. Despite the nodal structure for the large scenes, an ever-growing library of modules and buildings in the asset manager, and a way to maintain the sharp angles of its low-def models during rendering, these so- lutions were not enough when it came to addressing the biggest problem of all: the world cemetery, where 200 shots occurred. Like the floating village, the world cem- etery at the end of the movie was a single, huge set but contained dozens of detailed Hector the blue dragon proved difficult to model. He had to squash and stretch all parts of his body, including his eyes and teeth, without adversely affecting the skinning and blendshapes.

• Cover