Back to School
Issue: Volume 36 Issue 5: (July/August 2013)

Back to School

Success might not come easy to Pixar, but each year, as predictably as the first day of school, the studio that introduced CG features to the world scores another hit. Disney/Pixar’s Monsters University, the studio’s 14th animated feature and the fourth to revisit a previous hit, tallied $82 million on opening weekend to take the top spot at the box office. The combination coming-of-age, sports flick, and campus comedy achieved Pixar Animation Studios’ 14th-straight first-place opening.

In Monsters, Inc., Pixar’s fourth film, Mike Wazowski (Billy Crystal) was an assistant to his best friend, the company’s top scarer James P. “Sulley” Sullivan (John Goodman). Mike is a green ball on two legs with two arms and one eye. Sully is a big, tall monster with purple spots in his blue fur, horns, and a tail.

Monsters University takes place 15 years earlier. Mike, again voiced by Crystal, has a retainer and wears a cap. Sully, again voiced by Goodman, is thinner, has bed-head hair, and a big attitude. They are college freshmen. Mike arrives determined to become the top scarer. Sully comes with a reputation as a top scarer already. The two become instant competitors, with each trying to top the other. But, they both end up in the fraternity with all the misfits, the only one that would take them. How could Oozma Kappa possibly win the scare competition?

“One of the challenges of any sequel is that it’s creating a new game from an existing game,” says Director Dan Scanlon. “It’s like creating a new game using chess pieces and only those pieces. With a prequel, you have the added difficulty in that everyone knows how the story ends. So, you have to own that fact and use it to tell the story.”

In the story, we see the scarers in their formative youth and discover how they became the characters that starred in Monsters, Inc. They are teenagers in Monsters University, much less experienced than in the first film. Sully is an arrogant jock. He slouches during class, with a “who needs it” attitude. Mike is bookish, determined. He edges toward things, on the defensive against any blows that might come his way.

When the production crew at Pixar made Monsters, Inc. 11 years ago, it, too, was in its formative youth. The popular comedy was the studio’s fourth animated feature. Monsters, Inc. had one hairy character – Sully. The film met a CG milestone when Boo, the little girl in the loose T-shirt, touched Sully’s hair. Now, many of the students matriculating at Monsters University are hairy monsters, monsters in costumes, and hairy monsters in costumes that hug, sit, wear backpacks, and carry books.

“With every Pixar movie, we have some technical achievement,” Scanlon says. “On this film, one of the biggest challenges was how big the film is.”

Monsters University has approximately 400 characters, of which 25 percent have hair, often hair all over their bodies. While Boo’s T-shirt was the only costume in the original film, Monsters University’s characters wear 127 garments. Eighty-nine percent of the shots use a simulation to move something – hair, cloth, grass, flags, pages in textbooks, and so forth.

Fortunately, the production side of the movie-making team could tell Monsters University’s story using new “chess pieces.” “It’s a prequel, and we had to respect the design of the original film to fit into that world, but we wanted to use state-of-the-art technology,” says Sanjay Bakshi, supervising technical director. “We built a huge population of monsters and made sure some could be procedurally animated in a crowd pipeline. We had a new simulation pipeline on this film. We upgraded where it made sense. We used global illumination for the entire feature, which we hadn’t done before.”

PIXAR ARTISTS modeled and rigged 400 unique monsters from seven archetypes.

On Campus

Monsters, Inc. Director Pete Docter had wanted to populate that film with furry creatures, but he had to settle for one hairy monster. “He wanted a Muppet kind of world with funny, diverse characters,” Bakshi says. “We couldn’t achieve that then. Sully alone was a huge leap forward. But on this film, we didn’t impose any limits on the number of furry characters.” Or, on the number of arms, legs, eye stalks, horns, and so forth.

Character Designers Ricky Nierva and Jason Deamer, who had both worked on Monsters, Inc., started with subtle changes to the original models for Mike and Sully that helped them look 15 years younger. Then, they designed the other members of the Oozma Kappa fraternity. “We made Don Carlton, the mature student, a dragon because he’s dragging all that outdated stuff around,” says Deamer. They also gave him a batwing moustache, dry skin, and a polo shirt that he had washed too often.

For Jason, who is a mysterious vagabond character, they connected his arms between his legs. “It makes no sense, but in a great way,” Deamer says. “He’s purple, Muppet-like, and perfect.”

Scott Squishy Squibbles is moldable. “He’s underdeveloped, undeclared, a blank slate,” Nierva says. “For him, we thought about pliable substances. He has childlike proportions and five eyes.”

The most interesting and difficult character to design and rig, however, was Dean Hardscrabble (Helen Mirren). “In the 15 years I’ve been with Pixar, this was the most challenging character I’ve ever been part of,” Deamer says. “On the one hand, Dean Hardscrabble needed to be graceful and scholastic. But, she was the most successful scarer of all time and poignant in her criticism of Mike. Those ideas didn’t reconcile well.”

For a year, the designers worked with a male version of the character. But, six weeks before the character was due in shot production, they decided to change the creature’s gender. “It was like putting lipstick on a pig,” Deamer says. “Nine of us sat in a room and tried everything you could imagine: butterfly, moth, bat, crab, owl, spider, scorpion. We had a bird-like design that worked for a dean, but not the most successful scarer of all time.”

Then, a local insect dealer showed them a giant centipede.

“It is the creepiest thing ever,” Nierva says. “Dangerous, yet beautiful. Its legs create this elegant sine wave. We brought an expert in who deals with venomous snakes and poisonous spiders. He put on leather gloves and held this thing with long metal tongs. He said if you got bitten, you wouldn’t die, but you would wish for death. We had this thing in a Tupperware container with a little lid with tape on it in some poor coordinator’s office.”

Using the scary centipede as reference, the designers incorporated elements from horned dragons, gave her wings, and a designer garment. “She could afford the best,” Deamer says.

Rigging Dean Hardscrabble was a challenge, as well. “She is elegant and powerful,” Bakshi says. “She’s subdued and disciplined and scary in a professorial way, and then all of a sudden she flies out of a scene or uses her body to climb up a pole like a creepy but elegant centipede.”

Modelers working in Autodesk’s Maya and Mudbox incorporated her wings into her Coco Chanel-like jacket. “We don’t know she has wings until they burst out of her body like batwings,” Bakshi says. “We had to figure out how to collapse them because they’re always integrated into her body and jacket.”

The animators didn’t need to control each leg, however. When they put her on a path, a procedural system within Presto, Pixar’s proprietary animation system, would determine where each foot would go depending on the speed they set.

DEAN HARDSCRABBLE was the most difficult character to design and rig.

Student Bodies

Once the characters in the main cast were well under way, the artists moved on to the ensemble cast, a process that extended through two and a half years.

To create characters, the team initially used a system in which they rigged model parts in isolation and plugged them together, as if they were building LEGO models.

“At the time, we felt that was really clever,” Bakshi says. “But, people could see that the limbs we stuck onto a character weren’t really integrated that well, even with a fillet. We wanted to make characters with smoother lines that felt more organic.”

Knowing they would need to create a university filled with students and teachers, yet maintain a design consistency with the first film, the modeling and rigging teams began by looking at the variety of monsters in the original film. They organized those monsters into seven species, and then added controls that made it possible to vary the models and rigs for each.

“We had seven archetypes,” Bakshi says. “We would make a neutral shape, a slug, block, or pill, for example, and push parameters to vary the space between the eyes, how the tail tapers and its relation to the body, and the overall proportions.”

Variations in the overall proportions might include head and body size, limb, tail, and eye-stalk length. “The controls are essentially rigging controls, but for variation not animation,” Bakshi says. “Anatomically, the characters in each archetype are the same with the limbs, tail, and eyes integrated into the body. But, we could push the shapes around and add horns, fangs, wings, scales, fur, and other variations. Then, we could change shaders and colors, and make them smooth, wet, dry, rough. We built a huge population: 400-plus monsters.”

New tools within Presto helped layout artists and animators cope. “We have, on average, over 25 characters per shot,” Bakshi says. “The tools team did awesome work. We have an instancing technique so if a layout artist or animator had, say, 50 shots open with Sully and Mike in each, you could load all 50 shots and they would use the same memory as if the two characters were in one shot.”

The system also analyzed the tools as the animators used them. “We use the term ‘munging’ for a parameter like the one that moves the jaw up and down,” Bakshi says. “Often a program does that. So, if an animator mungs the avar [control] for jaw up and down, the system measures how long it takes to compute and execute, profiles what in the process takes longer, and then optimizes the computation. That was a big effort the tools department did for us.”

Animators could create performances – walking, flying, standing, clapping cycles, for example – for one character in a species, and then retarget that performance to all the characters in the species. “The system would measure the limb lengths and scale the animation to fit,” Bakshi says. “When we populated a crowd, we could apply the animation cycles to the crowd, offset them, and do procedural tricks. We had a ton of procedural animation, but it all originates with hand-drawn animation.”

Layout artists populated the scenes with crowds using tools developed originally for Brave that Technical Director and Crowd Supervisor JD Northrup integrated into the new Presto animation software and extended to manage the multiple species of monsters in this film. Northrup also developed tools within Presto that the layout artists could use to populate the crowds, place them, and see a lightweight representation for the first time.

“When you have crowds, they are basically like a set piece,” Bakshi says. “The goal was to have the layout shot composition persist through the pipeline. When these procedural characters got too close to camera or needed to react to something specific, we could parse them into a project for hero crowd animation. But, we always strived to use as much of the crowd animation as we could. Seeing the crowds and having them consistent at every stage in the pipeline really helped Dan [Scanlon] and the creative leads.”

RIGID-BODY dynamics moved the grass underfoot; FizT moved Sully’s hair.

Ubiquitous Simulation

All those characters – some furry, some with costumes – presented simulation challenges, as well. “We have simulation in 89 percent of the shots,” says Christine Waggoner, simulation supervisor. “Simulation is now ubiquitous. We had hair, cloth, grass, trees, ropes, paper, books, stuffed animals, newspapers, banners, bushes, backpacks, zipper tabs on backpacks, keychain on a backpack, skateboards, Frisbees. We had lots of beds in the movie, so we simulated the blankets, sheets, and pillows, and helped Sullivan lift a mattress to get to a dog. We even simulated Mike’s hat a little bit.”

An effects team created water, fire, explosions and other largely particle-based simulations with Side Effects Software’s Houdini. For the simulations relating to characters and objects, Waggoner led a small team of eight simulation artists who used the studio’s proprietary software.

Pixar scientists developed the studio’s first version of a hair and cloth system for Monsters, Inc. (see “Monster Mash,” October 2001). The simulator, dubbed FizT for cartoon physics, grew through collaboration among three senior scientists at Pixar: Michael Kass, the late Andrew Witkin, and David Baraff. While at Schlumberger Research, Kass and Witkin had published a landmark SIGGRAPH paper in 1988 titled “Spacetime Constraints,” in which they described animating a Luxo lamp using physics rather than keyframes.

Witkin and Baraff began working on cloth simulation at Carnegie Mellon University in 1992, and while there, developed Maya Cloth for Alias|Wavefront. All three scientists would join Pixar.

At Pixar, Kass developed a cloth-simulation system for Pixar’s Oscar-winning short film “Geri’s Game” (1997), and that was the launching point for new technology developed by Witkin and Baraff. They began working on the code base for the new simulator in 1998. The result was FizT, which solved a historic problem with cloth intersections. The three scientists received Scientific & Engineering Academy Awards in 2006, and Pixar patented the technology.

For Monsters, Inc., FizT managed Boo’s T-shirt and the 2,320,413 hairs that covered Sully’s purple-spotted, eight-foot-tall body. The slimmer, younger Sully now has 5.5 million hairs. The latest evolution of FizT, now upgraded and implemented within Presto, handled the dynamics for hair and cloth for Monsters University, as it has in every film since Monsters, Inc. except Brave.

“We didn’t have the problem with curly hair that they had in Brave, so we chose to use FizT for hair and cloth,” says Waggoner. “It simplifies the pipeline and it made sense from a legacy standpoint to have the continuity of technology.”

Because the characters didn’t wear the long, flowing garments that Merida and her mother wore in Brave, it wasn’t as necessary for animators and simulation artists to work together on this film. Typically, the simulation artists for Monsters University would run the dynamics on characters already animated. New to this film, though, was a representation of a character’s hair rendered in hardware.

“Animators on the original film could only see Sullivan’s skin,” Waggoner says. “But the composition looks different when you have voluminous hair. So, we collaborated with the animation department to give Sullivan a GPU groom.”

Under Foot

One of the more innovative techniques developed for Monsters University, according to Waggoner, was a vegetation-simulation system that managed the dynamics when hundreds of students trample the grass in the campus lawns. The proprietary system uses the open-source ODE engine within the pipeline for the rigid-body simulations.

“We created a technique that deforms the grass at render time,” Waggoner says. “We didn’t do a traditional curve simulation that you might think we would do because of the vast amount. To write out caches for that amount of data would be too heavy. So, we feed it in using a rigid-body simulation that drives a deformation at render time. It was a relatively innovative idea.”

The simulation artists start with a grass model created in the sets department, and specially rigged characters. “The grass model is a procedural system that renders curves that represent the grass,” Waggoner says. “The characters have rigs with simple spheres that interact with the grass. The spheres, sized to fit each character, are part of the rig in the sense of saying where the feet are. We use them as input, as colliders in the simulation.”

“When we run a rigid-body simulation,” she continues, “we tell it where the ground is and where the character is walking. The spheres fitted to the characters’ feet emit a trail of special spheres. These spheres have timing that allows them to float up at a certain velocity.” During rendering, the grass deforms according to the animation of these special spheres, which creates the appearance of grass coming up slowly over time behind a character’s footsteps.

The simulation artists created another interesting technique, this one to animate the pages in books. “A book has a lot of pages, so there are a lot of collisions,” Waggoner says. “People might think we’d use a cloth simulator, but very few simulators, even our proprietary simulators, can handle that amount of collisions. We used our rigid-body pipeline –­ the same engine we used for the balloons in Up. I think that might be surprising.”

To move the pages in the books, the artists created sheets of paper with cubes – that is, rigid bodies – linked together with joints. The rigid-body engine then moved the cubes based on parameters that specified joint velocity and stiffness. “I don’t remember how many rigid-body pages there were in a book, but we rendered 100 pages,” Waggoner says. “So, it was a large number.”

NEW, “SMART” raytraced lights made lighting easier and global illumination possible throughout.

Lighting and Rendering

One major change for the studio happened in the lighting department. Monsters University is the first film at Pixar to use raytracing throughout for rendering. “We felt RenderMan was ready, and we felt we could help the RenderMan team by using it in-house and pushing and collaborating with them, and improving it, as well,” Bakshi says. “We wanted to raytrace everything and simplify the lighting setup. Our films have looked visually rich and the lighting looks beautiful, but it was the lighting artists’ job to imitate the effect of global illumination. That imposed a complexity in the lighting setup.”

To reduce that complexity, a team of nine researchers, technical directors, and artists created a physically plausible, energy-conserving lighting system with importance sampling. Technical Director Christophe Hery, who had received a Scientific and Engineering Award for point-based rendering, was the driving force behind the new raytraced lights and global illumination system. Included in the system is a method he devised for raytracing hair with importance sampling.

With the new system, rather than dozens of point lights, the artists use any of six “smart” lights to illuminate a scene quickly and accurately. Three are shapes: square, disk, and sphere. Three represent large-area lights: sun, window, and sky. In addition, the artists could apply maps to the lights – even paint on them to direct rays to particular areas or colors.

“The goal wasn’t to be more efficient, although we were,” Bakshi says. “The goal was to simplify the lighting task. It came at some cost – a technological cost to develop the software and lights, and the cost of raw computing power.”

To render Monsters, Inc., Pixar’s renderfarm had 3,500 processors, a huge number at the time. Eleven years later, to render Monsters University, Pixar doubled the size of the renderfarm used on Brave to 24,000 processors. “Once we said that’s what we needed, it was such a big investment that we had to stick to our render-time targets,” Bakshi says. “We kept the sets not too geometrically complex. We counted the number of hairs, and if there were too many, we’d figure out how to make the same look with fewer number of hairs. To fit within the renderfarm capacity, we imposed this rigor on all the assets we built. And, we did a lot of optimization along the way. We had to make sure we could render the movie.”

Bakshi calculates that it took 100 million CPU hours to render the film, with each frame taking, on average, 29 hours to render. The result is a prequel that is orders of magnitude more visually complex that the original.

Pixar has always been a measure of the state of the art of computer graphics, and Monsters University is no exception. As with any good prequel, everyone knows how this story ends. It ends with Blinn’s law. Attributed to computer graphics pioneer Jim Blinn, it states, “As technology advances, the rendering time remains constant.”

A completely tricked-out college campus. Four hundred students. Hair. Raytracing. Simulation. Global illumination throughout. 24,000 CPUs. Amazing.

Barbara Robertson is an award-winning writer and a contributing editor for CGW. She can be reached at BarbaraRR@comcast.net.