The simulations took on average about 30 minutes each to calculate. This may not seem like a very long time, but since an innumerable number of tests had to be conducted to get a satisfactory motion, we had to have many computers working at the same time, each solving a different simulation, with varying settings. Integrity errors became another problem as many of our initial tests continued to crash. Upon cleaning up the mesh a little more, running STL checks, and tweaking the forces a little longer, the errors disappeared. Yet another problem occurred when we liked one simulation and the motion it created, but we wanted to push one of the pieces a little more in another direction, since changing a variable ever so slightly would create a completely different result the next time. Therefore, we didn't have much control over changing the simulation we ended up with, but rather a lot of control over creating it.
Once we had simulations we were pleased with, we had to texture map the animated mesh with the treatment Brad had come up with. This originally seemed like it would be a major ordeal because fully texture mapping the entire car, with animated textures for only 3 shots would have required an immense amount of time. Therefore we looked into other alternatives. Luckily, we realized that we could actually 'camera map' the first shot of the peeling sequence with a still of the first frame of the treated footage, and get away with it, since we never saw any other angles of the car except the front in that particular shot. 'Camera Mapping' refers to mapping an object with a texture projected from the view of the camera. Since the motion and staging of our 3D car matches that of the live action, and the movement the real car makes is not a drastic change in perspective, it is therefore possible to use the original shot as a texture. Also, since we don't ever see the other side, or back end, of the car for which the texture doesn't exist, this is once again an ideal situation for camera mapping. For the second shot of the top view of the car, we simply had to use the first frame of the treated scene as a texure as well, without any problems since it is a direct overhead shot without any significant perspective change. Finally, for the third shot, we could once again use a camera map because even though there is a severe change in perspective, the pieces have almost all but peeled off by the time the car begins to turn. The other thing to mention is that the underside of the mesh
was textured with a much darker tone in order to create a definitive contrast between the outer and inner part of the texture.
Finally, the lighting and rendering phase began (although there was no lighting involved since everything was rendered with self illumination in order to maintain Brad's treatment). Rendering was broken up into numerous layers such as background, foreground, shadows, etc, in order to give the compositor additional control over color correction, which proved to be useful during the online session with the agency. In addition, a GI pass was created to simulate the shadows that should appear where the peeling texture makes contact with the real car. Everything was re-rendered at the original resolution at which we received the files and, despite the high resolution involved, renders were fast as a result of zero lights and reasonable
The final spot was composited using Combustion, and then sent back to Toybox for dumping back to film. There were several transfer tests done first, since there sometimes tends to be dramatic changes in color going from file to film. The final 61 second spot began airing in Canadian theatres in March, 2002. There is also a 30 second version airing on Canadian
and American television networks.
The software used to create "Toyota Matrix" was 3dsmax, Simcloth, US Animation, Flash, After Effects, and Combustion.
View the final movie here