BEN GROSSMAN: Two days before the shoot, the spot was called "Sharks." It was completely previs'd, and was supposed to have a completely different concept, drawing attention to a completely different feature of the car. Once we finished brainstorming a new concept with the creative team at Publicis and director Michael Haussman, we had to restructure the team and the approach in just a couple days. Where the old spot was supposed to be in the sunlight, and involve CG characters, the new concept was in the rain and required a giant matte painting/CG shot at the end. So we went from a character-animation oriented team to a more FX oriented team and worked out the details of a whole new slate of production equipment that would give us rain interaction on tight shots, wet-downs for the road, and a set-piece for the bridge destruction at the end. On the comp side, we knew we were going to have to roto not just the cars, but also all the skies, and even lots of parts of the car in order to add in details that would exist in a rainy environment but couldn't be shot in broad daylight, like headlights and reflections.

What tools did you use to execute the effects and construct the rainstorm, and why did you choose them?

Given the tight turnaround, we figured the best approach was to create a short stretch of road covered by three rain towers for the tight shots of wheels and whatnot, and use CG rain for all the medium and wide shots. In the end, we added CG rain to every shot, but the practical interaction we got from the rain towers was something we couldn't have done without. Instead of trying to run the rain out of a typical 3D application with the traditional à¼ber-long render times and motion-blur issues, we used a Trapcode plugin called Particular [www.trapcode.com], which is a 3D particle system you can use directly in After Effects. You can get really nice motion blur, depth of field, and a few million raindrops for as little as 15 seconds a frame. We also used it to create the distant volumetric rain coming from individual clouds.

Of course, much of the footage was shot car-to-car or from a SpaceCam, so we imported 3D camera match-moves into all of the comps so that the rain all moved properly with the camera. We also did match-moves for the cars so that we could attach volumetric headlights to them that were generated from another Trapcode plugin, Lux.

All of the sky replacements were composites of various clouds taken from our image library and stitched together to make something new and good-looking.

Normally, a spot like this would require a lot of CG artists to crank out a lot of FX elements, but we managed to set it up so the compositor, Josh LaCross, generated almost everything, which made it a lot faster and more change-friendly. There was only one shot in the commercial, the end shot of the bridge destruction, which required a CG element, but even the falling rocks bouncing off the rock wall and road in that shot were done in-comp using Particular.

How does rotoscoping with Motor improve on your existing methodology?

We knew that a huge roto team would have put us over budget, or would have taken away from the quality of the other FX, so we took a wack at using Imagineer's Motor product [www.imagineersystems.com], which was in beta at the time. Traditionally, if you want a matte for a car, you roto a matte for the car. Then if you want a matte for the windshield, you roto a matte for the windshield. Repeat for headlights, fog lights, wheels, etc. With Motor the concept is different. The Planar tracker tracks an object like a car, and then attaches the roto shapes to it. Because it accounts for changes in position, rotation, scale, and to some extent, perspective, it does a much better job adjusting your shape to match the object than if you attached the shapes by hand, or with more traditional methods of tracking.

We did a comparison rotoing a car in a 100-frame shot by hand, and we ended up with about 88 keyframes. When we did it in Motor, because the program essentially "neutralizes" the camera move and compensates for translational changes in the object, we ended up with only 4 keyframes – one on either end and a couple in the middle.

The best part though, is that once you've processed a good track for an object, you can create any number of additional mattes on the same object "for free". In any other program, you probably need to make a new shape, say for headlights, with its own 88 keyframes. In Motor, you just make a new shape for each headlight, attach them to the master shape, and maybe set a couple keyframes, but you're pretty much done. Because car commercial people are nuts – and they have names for every little detail of a car and they are always wanting each of them to be groomed and toned separately until they make the product look like a DeBeers diamond – it's damn useful to be able to bang out 10 different mattes for different parts of the car in just a few minutes.

Because some shots and some elements are more receptive to the planar tracker than others, the time savings can vary quite a bit. When it's not helping much, it's at least as fast as traditional rotoscoping. When it's dominating the shot, it can save days.

We were sitting in the Flame bay with the clients, and at the last minute they suddenly wanted to know if there was time to color-correct the grill of the car separately to get in more detail, and we all instinctively said "No," because we were thinking it was gonna take too long to roto. But, remembering we had processed that shot through Motor, we called our Motor operator, Tim LeDoux, upstairs and said, "We need a grill matte." We had a grill matte in minutes. The agency team all wanted to trek upstairs to see how we did it.

Were there any compromises involved to meet the accelerated deadline?

We expected to have to compromise on a lot of things, but we ended up delivering on everything – which is a bummer because I'm sure next time they'll slash the budget and double expectations.

What else is new in your pipeline?

Right now, we're rolling out 3ds Max [www.autodesk.com], in addition to our existing 3D apps, with several different render options. We tend to use the "Noah's Ark" approach to our pipeline so we aren't constrained by the limitations of any one piece of software. Plus, we're a beta facility for a lot of different stuff, which always keeps things exciting.