To animate the characters, animation supervisors used traditional key-frame animation, combined with a great variety of simulations and dynamics.

Once the design phase was completed, the team started working on the characters. First, an artist sculpted character study maquettes. These were then used as a reference to model the digital characters. "None of them featured any real modeling or rigging challenge, except for the dog and the World Eater," Brack notes. "The former had to be able to stretch and squash in any direction, a la Tex Avery, which presented certain difficulties. As for the World Eater, the challenge was its sheer size: 450 feet (150 meters) tall! It was going to appear in wide shots and in tight close-ups. At one point, the characters manage to climb on its back, and the dragon's body turns out to be so huge that it becomes the whole set in itself! That presented quite a challenge in terms of shading. We didn't want to have one shader for long shots, another one for waist-up shots, etc. So, we managed to develop a single shader that worked from any distance."

Layering Up Animations
When time came to animate the characters, animation supervisors Kyle Balda and Laurent de la Chapelle and their team used traditional key-frame animation, combined with a great variety of simulations and dynamics. "The main animation was all key-framed, but there were many secondary animations to take care of," Brack explains. "Zoe's braids, Lian-Chu's ponytail and other hairs were either hand-animated with dynamics added on top, or a dynamic simulation that we manually fine-tuned. The directors didn't want these animations to draw unnecessary attention. Fur simulation also had to remain subtle. All the characters' costumes featured some kind of fur, and the dog was, obviously, all fur. We used a proprietary fur system to create these different looks."

The team also combined key-framed animation with soft body dynamics for a giant caterpillar-like dragon. The body fat had to bounce and wiggle in a realistic way as the creature moved around. RealFlow was employed to generate a fluid simulation for the sequence in which the same nasty dragon throws up a deadly substance.

Animation was not only limited to the characters, as the environments themselves required a great deal of rigid body dynamics simulations. "I don't think there was one single set in the movie that we didn't destroy in one way or the other," Chauffard observes. "The final sequence takes place in a ruined city that is being destroyed as the action progresses. We built about 50 different buildings out of fully textured blocks, so that when a piece exploded, it looked fine from every angle. The same blocks were used in another sequence to create thousands of ruins that are magically hanging in midair. Those individual pieces were attached to particles that were then animated dynamically, creating this eerie floating movement. The sky was a 360-degree cyclorama that allowed us to point the camera in any direction. The clouds nearer to camera were volumetric renders."

When a scene combines multiple character animation, fur simulation, dynamics simulation and a complex environment, render times can become intimidating.

Render Issues
When a scene combines multiple character animation, fur simulation, dynamics simulation, a 450-foot dragon, and a highly complex environment, render times can become intimidating. However, Mac Guff Ligne managed to keep those processing times under control. "When we started this project, I told everyone that we shouldn't have any shot requiring more than two hours of render time per frame," Brack recalls. "That was our limit. So, we developed lighting set-ups that enabled us to have relatively constant render times whatever the image was. When we noticed that a shot required much longer, it meant there was a problem somewhere. It could be as simple as a reflection on the corner of an eye -- it didn't bring anything to the shot, but it was very costly in terms of render time. So, we always took the time to identify these unnecessary details and get rid of them in order to bring the shot down to a reasonable processing time."

Keeping control of the render times was of paramount importance for the team to meet the deadline: the whole production lasted less than 18 months... "We worked a lot in layers as a way to optimize the whole process and have a better control on the final result," Chauffard concludes. "We had one layer for the characters, one for the hair, one for the environment, etc. Each layer had its own render time, and the total render time couldn't be over two hours per frame. It meant that compositing played a key part in putting those shots together and in creating the final look. Then again, we couldn't have done it without the very precise technical preparation that took place in the early stages of the project. We worked really hard early on to fix what needed to be fixed. There was no 'We'll figure it out... ' In the end, we managed to meet our deadline, and we still produced some remarkable images."

Alain Bielik is the founder and editor of renowned effects magazine S.F.X, published in France since 1991. He also contributes to various French publications, both print and online, and occasionally to Cinefex. In 2004, he organized a major special effects exhibition at the Musée International de la Miniature in Lyon, France.