The team started with a high-detail Lidar scan of the set. The complex geometry was then extensively optimized and rebuilt. Displacement maps were extracted from the fine-grained bark detail, and a library of trees was created to be fitted onto the stumps in the set. Lower-detail tree models were created for use in the mid distance, whilst the far background was finished with a high-resolution digital cyclorama created in a combination of Photoshop and Stig, part of a proprietary 2.5D environmental toolset. For shots looking up into the canopy, the team modeled complex patterns of branches and added thousands of leaves with a modified fur system.

Meet Grawp, the Teenage Giant
Development on Grawp started as soon as Double Negative joined the project. The CG character was a major challenge on several different levels; he had to look real, but also needed to deliver an emotionally complex performance. Extensively previsualized, the Grawp shots were filmed using an innovative virtual set system that allowed the director and crew to "see" the giant on set composited into the live video feed from the main camera. It enabled them to produce tightly choreographed camera moves and performances that responded accurately to the giant's notional position on set.

"The system was created in collaboration with Lester Dunton of Joe Dunton Cameras (JDC) and third party developer Olaf Wendt of United Image Systems," Franklin explains. "Basically, we took realtime data from a motion-encoded Supertechno 30-foot camera crane, and fed it into Autodesk Motion Builder where it was used to drive a CG camera on a model of the actual Forbidden Forest set. Custom tools were developed to allow the Motion Builder set to be registered quickly in 3D space with the real set. Senior programmer Oliver James devised a workflow to import Maya generated previs animation into the system, where it could be manually cued to respond to the movements of the camera crane. The on-set system consisted of a PC laptop running Motion Builder. It fed the output of its video card to a small vision mixer where it was composited in realtime with the feed from the videotap on the main 35 mm camera. The composited output was then relayed to the camera operator who could see the 3D animation's position and action overlaid onto the image of the set."

Using a high detail Lidar scan of the Forbidden Forest set, DNeg built forest canopy extensions. For shots looking up into the canopy, the team modeled complex patterns of branches and added thousands of leaves with a modified fur system. 

For the character itself, the team developed a very detailed muscle and skin simulation. They added many secondary dynamics to give jiggles and wobbles of the correct scale as Grawp lumbered around the Forest clearing. "We used an advanced facial rig that was specifically designed to give our animators very precise control over the facial shapes," Franklin notes. "One of the rig's best features is that it can take input from almost any source, as we have worked out how to feed in data directly from a number of motion capture systems. It allows us to use that kind of data as a basis for a performance if needed. At the outset of animation development, we used a new motion capture system developed by Image Metrics to capture the facial performance of actor Tony Maudsley who was cast as the reference for Grawp. In the end though, due to changes in the sequence during postproduction, we eventually keyframed all of the animation."

Shading supervisor Graham Jack developed a sophisticated approach to simulating Grawp's skin. "Initially, the skin work started with fast shadow map based scattering, but as look development progressed, it became obvious that this method was limited by the fact that it only supported spotlights, and that it only simulated single scattering," Franklin continues. "To overcome these problems, we added Pixar's point-based subsurface diffusion approximation. This uses a multi pass approach which bakes surface illumination in the first pass to a point cloud, computes the subsurface illumination on the point cloud, then reads it back in for the beauty pass. Various tools were written to help fit the technique into DNeg's existing shader setup. In theory, using the point cloud based scattering would have circumvented the need for the shadow map approach, but in practice, the multipass approach didn't give enough detail, so the final result used a combination of the two approaches. In addition to the scattering, the shader also made use of the new point cloud-based color bleeding in PrMan 13. This allowed the very fast calculation of indirect illumination, and really helped in the areas around the ears and eyes."

Skin specularity was catered for by three separate specular and reflection lobes, each of which could be texture mapped. This allowed very careful balancing of an overall specular sheen, glossiness from surface sweat and separate glossiness from liquid on the skin surface on the lips, eyes etc. Numerous displacement maps, representing different facial expressions, were blended together in the shader to achieve the final displacement. The blending of these maps was controlled by the character's animation rig. They served to augment the facial animation system with details such as wrinkles.