The film Cars had entire sequences where the animation and rendering of tens of thousands of cars was required. This presented a unique challenge, especially in the case of the racetrack stands. We needed a way to render the sheer volume of characters in the stands, and also to animate each character so that it looked unique and non-cg. There were literally hundreds of car variations created to add interest and make the world more realistic. We created a special shading model for rendering a massive number of cars, which we called “Shrinkwrap” technology. We also created a procedural model called the ”CarCrowd” model which would allow us to deal with animation on a very large scale. In addition, it was necessary to create software that made applying animation to a large number of characters fast, simple, and robust.

A Parking Lot of Variation

We were tasked with developing hundreds of diverse background characters to create visual interest. We implemented a collection of model and shading variants which allowed us to use a small number of base characters to create a large palette of cars with different shapes and sizes. We also built a variety of hubcaps, accessories, props, shading/paint passes, and shaded headlights that were added to each character. By modifying each of these attributes we were able to construct a huge parking lot full of characters, from which cars were then cast. Many of these options were built into the Shrinkwrap cars so that the same variation seen in hero cars could be seen in crowds of cars in the stands and elsewhere.

From Boxes to Cars: Rendering

The main idea behind the Shrinkwrap model was to use very simple geometry to represent the car, then use textures to displace the geometry and assign materials to specific sections of the cars. These Shrinkwrap cars were geometrically very simple: a box, 4 planes for wheels, and optional planes for any hats, flags, or pendants the car might be holding on its antenna. The shader used several texture maps, including maps for displacement, materials assignment, and special maps for wheels, hats, flags, and eyes. Because there was a set number of textures used on all Shrinkwrap variants, the number of texture reads was minimal. The Shrinkwrap shader used textures generated from full-res renders of the actual hero cars, and made it possible to render hundreds of thousands of Shrinkwrap cars with ease.


Creating Chaos out of Order: Animation

In order to create animation for such a large number of characters, it was imperative to have a mechanism to apply a small amount of animation information to a large number of cars. This was done procedurally in the CarCrowd model. The model for the racetrack stands contained a number of individual actors, each of which was simply a type of Shrinkwrap car. This way a single model could contain hundreds of cars without causing a lot of overhead. There was a special version of this CarCrowd model used as a prototype for animation. Containing 40 to 50 cars, this proto-model was animated using our animation application tool. The stands model would receive its animation from this proto-model, as each car in the stands would randomly choose an actor in the proto model and play back its animation (with some offset). This gave the impression that all the cars in the stands were performing unique actions even though they were in fact only reading from about 40 or 50 different animation loops.

Under the Hood: Animation Application

Our tool for applying animation was written to be easy enough for animators to use, yet powerful enough for crowds TDs to perform complex tasks. Animators created distinct animation clips for short actions, such as cheering1, cheering[n], booing1, booing[n], as well as transitions to and from different animations, such as cheering-to-booing1, cheering-to-booing[n], etc. These animation clips were then placed in a state machine which indicated the start state, the order of animation clips, transition information, triggering transitions, and weights of clips within states. Multiple state machines were created for different types of actions, such as cheering, keep-alive, etc. We would then apply a given state machine (or several state machines with different weights) to a proto-model, giving us a range of animation on the proto-model actors with different offsets and scaling. Because the tool was only applied to a relatively small number of cars (less than one hundred), this was a fast and efficient way of animating over one hundred thousand cars transparently.

This tool was also used by animation to deal with shots containing a large number of hero cars, which would normally be hand-animated.