ffmov

Reality Check

The first CG cast of human actors stars in
Final Fantasy: The Spirits Within

By Barbara Robertson

All Images courtesy of Square Pictures, © Columbia Pictures.

Pick a science fiction film, any film. Alien, Star Wars, Men in Black, Fifth Element, you name it. Now, imagine the same film entirely created with 3D computer graphics-not just the visual effects, but also the people, cities, landscapes, spaceships, sun, moon, stars, every little bit. That exactly describes the movie Final Fantasy: The Spirits Within from Columbia Pictures and Square. The film, which was released July 11, puts synthetic human actors into roles that could easily have been played by real humans and places them in completely synthetic sets. Four years in the making, the film is one of the most ambitious computer graphics projects ever undertaken.

"The story could have worked as a live-action film," says Hironobu Sakaguchi, director of the film and CEO of Square USA (Honolulu), the studio in which a crew of around 200 people drawn from live-action films, games, commercial production, animation, and other areas crafted Final Fantasy. "We decided we wanted to use realistic computer graphics. Our goal was to present the audience with something they'd never seen before."

The decision makes sense given Final Fantasy's starting point as a role-playing game, and Sakaguchi's role as the originator of that interactive game. With worldwide sales of Square's nine-part Final Fantasy series totaling more than 26 million units, according to the company, the studio could count on a ready-made potential audience of game-players familiar with the fantasy themes, comfortable with computer graphics characters, and eager to see what happens next. Sakaguchi did not choose to duplicate characters from the video games, though. Nor did he choose a storybook style of 3D animation, such as that in the films Shrek and Toy Story. Final Fantasy has an original screenplay written by American screenwriters Al Reinert and Jeff Vintar, and a style all its own.

Square's filmmakers describe that style as "hyper" real not "photo" real, noting that they tried to achieve a heightened reality in the film, not a replicated reality. Final Fantasy doesn't look exactly like reality, but it doesn't look exactly synthetic either. It could almost be a live-action film that was altered for effect. The people at Square believe it could change the way films are made in the future. "The existence of the film itself is a great demonstration of what computer graphics can do," says Sakaguchi, who feels now that people know it's possible to create human characters on screen, the door is open to new ideas.

"It's a shot across the bow in terms of cinematography and traditional animation," says Gary Mundell, CG Supervisor. "We can do anything we want now. In this type of filmmaking, there are no limitations on special effects. There are no limitations on scenery. We can put a human character into any situation and make it do anything. There is no story that cannot be told."

Square can use words like hyperreal, but it is the photorealistic look of the film's human characters that is, arguably, the studio's most stunning achievement. Indeed, several months before the film was released, the men's magazine Maxim pushed aside flesh and blood women to put Final Fantasy's star Dr. Aki Ross, wearing a bikini in a seductive pose, on the cover. And why not? The women on magazine covers will never be anything more than photographs to most readers anyway, and Aki is as beautiful as the rest.

Staging Emotions

To direct the cast, Sakaguchi started with detailed frame-by-frame storyboards, which were scanned and converted into a slide show on an Avid system. Then, once the voice actors had recorded the script, a layout team began its work. Using Alias|Wavefront's Maya running on SGI workstations, scenes were blocked out with simple 3D models of characters and "filmed" with virtual cameras. "We went to a '70s and '80s style of camera, using [virtual equivalents of] cranes, steadi-cams, and hand-helds," says Tani Kunitake, staging director. "We wanted a cinematic style, not a game style."

The storyboards and 3D layouts were both sent to the animation and motion-capture departments. The 3D layouts showed the camera moves and blocking for the characters timed with the voice recording; the storyboards showed the subtext, the characters' emotions.

"You can't put expressions into a cubic puppet," Kunitake says, referring to the 3D layouts. And the voice recording used for the 3D layout didn't tell the whole story either. "Sakaguchi-san would discuss what the scene is about visually, the relationships between Aki and Sid, Aki and Gray," Kunitake says. "He would discuss an emotional path that would happen in a scene; how the characters feel. We would draw these things putting emphasis on the story rather than the dialog."

Small Changes

Trying to turn a 3D model into an emotional and realistic human character, though, involves a delicate balancing act: If the 3D character doesn't look human, it doesn't matter whether it moves like a human or not-and vice versa. But a change to the look can throw the animation off-and vice versa.

"They're both challenging," says Andy Jones, animation director. "We were always go ing back and forth with the character group about what was more real, the character's motion or look."

Jones decided to have animators concentrate primarily on the characters' faces and fingers, assigning lead animators to each character, and also decided to use motion capture technology to help the animators move the bodies. "Originally, we thought we'd capture 20 to 25 percent of the human movement and the rest would be keyframed," says Remington Scott, motion capture director. "We thought it would just be used for characters when they're running or moving in large ways."

Although it was effective for recording that type of motion, the team discovered that motion capture became even more valuable when characters were not moving. "When characters are standing still and starting to talk and relate to each other, there are incredible subtleties and nuances," Scott says. "These things are the essence of what it is to be a human, to move like a human." Slight movements that could change the mood of a scene, such as the way a person stands and balances his weight, or shifts around in a chair were difficult to animate but were relatively easy to capture. Thus, when the team began casting motion capture performers, they looked for acting experience. Tori Eldridge, an actress and dancer with a black belt in two martial arts performed Aki.

Square uses a 16-camera optical motion capture system and tracking software from Motion Analysis; each performer wore around 35 markers, five of which were in the chest area to capture breathing.

Proprietary software made it possible for Jones, Scott, and Jack Fletcher, voice and motion capture director, to look at the motion capture data applied to animated characters in Maya within a few minutes. "It wasn't real time, but it was quick and extremely accurate because we could see the characters with all the appropriate elements in Maya," says Scott.

Ultimately, Scott estimates, 90 percent of the body movements were based on captured data. That data was fed into an in-house system, which converted the captured performance into a form that could be used in Maya. "You can scroll a character back and forth in Maya and see the motion captured figure moving behind it like a ghost," Jones says. "If you see motion capture actions or poses you like, you can snap to them, or you can skip sections and animate those by hand."

The characters' bodies were all created in Maya using a representation of a muscle structure under the skin, and a rig that caused joints to move bones, which moved muscles, which moved the skin. "The body had to be anatomically correct," says Kevin Ochs, character technical director, "because we had to worry about how the clothes would move. To keep things efficient, however, the team worked with only as much of the body as necessary. "If you saw a character only from the waist up, he wouldn't have legs in that shot," Says Ochs.

Moving Faces

Once the body performance was finished, the animators began working on the characters' faces, which were animated entirely by hand. To help make the skin move realistically, the technical team developed a system that made it possible for the animators to give the skin a looser feeling than with shape blending alone. "Blend shapes stretch or shrink the skin from one point to another, which doesn't allow much leeway for creating looseness in the skin," says Ochs. "We created a deformational cage that allows us to represent muscle structure on the fly when the face starts moving, to give the skin an elastic feel." The technical team used the same cage, with minor modifications, for all the characters. Basically, the cage pushes and pulls on several key points depending on the facial expression, which causes the skin in those areas to move around the points rather than stretch from one point to another. "We isolated points that seem to be in constant motion while a person is talking," Ochs says.

To create wrinkles, the team decided to use animated displacement maps. "We realized that one technique was not going to give us the look we wanted," Ochs explains.

"It's extremely difficult to get the movement of the skin on the face to look real," says Jones, "especially with how close and unrelenting our cameras are. Like when a person frowns, little dimples and slight imperfections appear in the skin that are so difficult to get in a CG character. You can use animated texture maps, but if they don't perfectly match the motion and the emotion of the character, something looks off."

Lighting affected the look of the animation as well. For example, Jones explains, there's a tendency for the bottom of the trough of a wrinkle to be a curve rather than a hard line. The curve causes a reflection of light, which looks odd inside a wrinkle. Even though the team used techniques within Pixar Animation Studio's RenderMan as well as in-house tools to help solve that problem, it was difficult to make the wrinkles look real when they moved-and there was always another CG idiosyncrasy waiting in line.

"There are so many subtleties," Jones says. "We made a stylistic choice to have our characters look as real as possible and, given that style, to get as much emotion as we could out of them. Ultimately, I think the look side won, especially with faces. It's just so difficult to get the facial movement to look real."

Looking Real

To make the characters' skin look as real as possible, Ochs needed a method that would render quickly and adapt easily to various lighting conditions. First, he devised an algorithm that managed the consistent properties of skin, such as how it reacts to light. Then, the team wrote a RenderMan shader to create the inconsistent properties. By using texture maps and procedures, this shader created approximate representations of skin properties that varied in different areas of the body or as external conditions changed.

Shades of gray from black to white in a texture map used by the shader, for example, would indicate varying skin thickness to determine how much light would shine through backlit ear tissue compared to fingers. And, procedures in the shader managed such skin properties as colors and specularity. By varying the amount of specularity, for example, the team changed the look of the skin from oily to dry. "A character's arms would have less oil than its face," Ochs says. "In some cases, when they're running or under stress, we would kick up the specularity values so it would look like they were starting to sweat." Finally, other details such as freckles and moles were created with painted texture maps.

While the skin helped make the digital actors look as if they were made of flesh and blood, creating believable hair would be the team's crowning achievement-particularly Aki's hair. Using a proprietary system, animators could have a simulator automatically generate and animate Aki's hair based on her movements and settings they entered into the system. For example, an animator could choose an "Aki running" hair setting that gave the hair a particular kind of bounce and also specify a strong wind blowing into her face. Or, animators could use control hairs to create specific actions such as Aki putting her hair behind an ear. The simulator would then grow the rest of Aki's hair and move it to match the movement of the control hairs.

The initial hairstyles for the digital actors were all designed with the control hairs, which were spline-based curves that acted like hair plugs. The approximately 50 control hairs per head would be output from Square's system as RenderMan Ri curves, and during rendering would grow into 50,000 to 80,000 hairs, depending on the hairstyle. To shade the hair, the team created special algorithms. "The curve doesn't have a rounded surface, so we had to do a lot of little cheats to give a realistic look to the hair," Ochs says. "A lot of R&D went into how lighting hits it."

After it came to clothing, though, the animators got a break. When an animator finished a character's performance, the animation was brought into a costume file, which housed the character's body and clothes. There, the animation was attached to the body, and the body's motion would drive a proprietary cloth simulator. "The entire cloth simulation was done for this movie in about a year and a half with about four people who would set up the clothing and give it the properties it needed. It was a very streamlined pipeline," says Ochs. Because the team designed the software specifically for this show, they built into the program the features needed for the movie-leather, rubber, wool, stiff collars, shoulder pads, and so forth. "It allowed us to stay very clean and quick," Ochs says.
Being clean and quick was important: Although most astonishing, the human stars weren't the only animated characters in Final Fantasy. The animation team had to manage crowds of digital people in the cities, herds of ethereal phantoms, and various kinds of vehicles, ships, and jeeps that typify a science fiction movie. "Ev erything gets overlooked because of the humans," says Jones. "But we had to animate anything that moved and wasn't created with particles or effects."

Fantastic Effects

"We have craters collapsing, dust, clouds, fire, explosions, debris," says Remo Balcells, visual effects supervisor. "We even blow up a whole planet." Balcells' team combined techniques used in games with RenderMan shaders to create many of the natural effects. For example, to create fire, they used a billboard technique. "We mapped a single polygon with a transparent texture map that has the colors of fire on it," he explains. "And then using a Render Man shader, we animated the texture procedurally."

To create smoke trails, the crew used a similar technique. "In stead of a single rectangle, we use many of them mapped onto particles," says Bal cells. To create the illusion of self-shadowing columns of smoke, however, they mapped the shadows of simple spheres onto texture-mapped spheres. Spheres were also used for fireballs created from explosions. "We'd map the spheres with a volumetric shader and render them," he says. "It's not a complex technique, but it requires a complex shader."

As production progressed, though, the team began to rely on compositing for many atmospheric effects such as fog, which could be created more quickly with a compositing program than with rendering. "At the beginning of the production there was little emphasis on compositing," says James Rogers, compositing supervisor. "It was seen as something that could add the odd lens flare." By the end, many of the shots were composites with more than 100 layers. "A lot of people who came from live action like myself are used to working that way," he says. The team started with Avid's Illusion, then began working primarily with Nothing Real's Shake and Adobe's After Effects, with Alias|Wave front's Com poser and Discreet's Flame also playing a role. "We used anything we could get our hands on," says Rogers. "I think that we ended up with an average of 16 layers per shot. Early last fall, after he took on the job of CG supervisor, Mundell did an inventory of shots that had to be completed and realized there were far too many frames to render. By ad ding WAMnet service to the studio's 1000 Linux-based Pentiums, 300 SGI Irix CPUs, 160 Octanes, and 3 Origins, and by isolating areas that were taking more time than others and optimizing the workflow process, the team was able to finish three weeks ahead of schedule. "We found that Aki's hair was taking 20 percent of all our rendering time," Mundell says, for example. So they optimized the hair rendering and also changed such things as texture map sizes where possible. In terms of workflow, they tried to reduce the rendering times for shots in process and the amount of approvals needed for each shot. "Next time it will be a lot easier," he says.

Like many people at Square, Rogers and Mundell believe that the film is going to cause a paradigm shift. "Filmmaking has always been driven by technology, from black and white to color, from silent to sound," Rogers says. "I think this is one of those things. It's not necessarily the right way to make a film, but I think it's quite significant. Filmmaking is ruled by economics, and as the cost of technology decreases, this style of filmmaking may become more popular. Instead of flying a crew to location x you could create a virtual location. Instead of calling back actor x, you could do a cutaway shot in the computer in post production."

Recently, a Hollywood agent who wants to represent Aki approached Sakaguchi, a development he welcomes. "She could be hired for commercials, or other movies," he says. Thinking about those implications could put a regular human into hyperdrive. Would she appear as Dr. Aki Ross, the scientist, endorsing a product? Or as the actress who played Dr. Aki Ross? And who is that actress anyway? Is she Ming-Na, Aki's voice, Tori Eldredge who moved her body, the modelers who built her body, the TDs who gave her flesh and blood, or Roy Sato, the lead animator who gave her expression? Does she exist without all these people working in concert?

Whether Aki becomes a star in her own right (whatever that is), there's no question that photoreal or hyper real or simply realistic digital actors will appear in future live-action and animated movies in important roles. Perhaps this will open the door to actors and actresses for whom the camera is not a friend, allowing them to become stars by stepping into camera-friendly CG bodies and speaking through the lips of CG characters. It's easy to imagine all sorts of possibilities-movies in which one actor plays many roles voiced by multiple actors . . . or multiple actors playing many roles all voiced by one actor. In the end, will the digital actors be the stars or the people driving them?

It's possible that at this moment in time, the answers don't matter. What matters is that because of Final Fantasy, we are thinking about it.

Barbara Robertson is Senior Editor, West Coast for Computer Graphics World.

Computer Graphics World August, 2001
Author(s) : Barbara Robertson