ARTICLE BY JOHN DUNCAN

When visionary filmmaker James Cameron, effects master Stan Winston and visual effects executive Scott Ross founded Digital Domain in 1993, they did so with a specific and ambitious goal in mind — to create, from the ground up, an effects company that would be at the vanguard of computer graphics, a place where technology would not just serve, but actually inform the creative process. "We wanted to create iconic and memorable characters and creatures for pop culture, using new techniques," said Cameron, "much like the early days of Disney. For Disney, the core of his business strategy was to advance the art of cel animation. We wanted to do the same thing, advancing CG.

"But it didn't work out, at least not back then, because the first few shows Digital Domain attracted — even the shows that I was doing — worked against that strategy. True Lies and Apollo 13 were both 2D-comped. We had Titanic coming up, which was going to be a huge show, but it was going to be 2D-comped, too." To make Digital Domain the groundbreaking facility that he and his partners envisioned, Cameron began to think long-range, to conceive a project so cutting-edge that it would push CG technology — already advancing at a brisk pace — to its foreseeable limits. That project was Avatar.

Avatar is set in the future, on the distant and lushly forested moon of Pandora, where an earth-based company is mining an extremely rare and precious mineral given the colloquial name 'unobtainium.' As its name suggests, this mineral is very difficult and expensive to mine, largely due to the inability of humans to breathe Pandora's toxic atmosphere. The mining company has overcome that problem with a sophisticated avatar program in which the DNA of a human is combined with that of a member of the indigenous population — a ten-foot-tall, lean, blue-fleshed people called the Na'vi. The result is a hybrid creature, controlled by the original human via a link machine that connects his consciousness with the avatar's body. An avatar for scientist Tom Sully, in which the company has invested millions, is rendered useless when Tom dies in a robbery. To recoup their investment, the company tries to recruit Tom's twin brother, Jake (Sam Worthington), a jaded, wheelchair-bound former Marine whose matching DMA makes him the only fit controller for the avatar. Originally disinterested in replacing his dead brother, Jake reconsiders when he realizes that, as an avatar, he will walk again.

After a six-year trip aboard a spaceship, asleep within cryovaults, Jake and his fellow recruits arrive in Pandora and get their first glimpse of the moon's gloriously vivid landscape of 900-foot-tall trees, luminescent plants, and mountains of floating rock suspended by a forceful magnetic field. But for all of its beauty. Pandora is a dangerous, wild place, with wary Na'vi and ferocious predators just outside the fences of Hell's Gate, home base for the humans' military-industrial endeavor.

Jake is thrown into that dangerous world when, in his first avatar mission into the forest — accompanied by the avatar for Dr. Grace Augustine (Sigourney Weaver) and her science team — he encounters a thanator. Pandora's 'king of the jungle.' Chased by the predator, Jake becomes separated from his group and spends the night alone in the jungle, stalked by Neytiri (Zoe Saldana), a beautiful Na'vi woman who finally saves him from a pack of viperwolves. Back at Hell's Gate, the military commander Quaritch (Stephen Lang) orders Jake to exploit his fledgling friendship with Neytiri, to infiltrate the Na'vi and learn their ways so that the company might gain control of Pandora. What begins as an assignment becomes a journey of discovery for Jake, who falls in love both with Neytiri and the Na'vi culture. In the end, Jake — who has finally found something worth fighting for — leads the Na'vi in an existential battle against the human intruders, well aware that with their expulsion will go the scientific support for his avatar body, and thus his life with Neytiri.

Cameron wrote the Avatar treatment quickly, finishing it just before he left for his 1995 expedition to dive the Titanic and capture footage that would make its way into his 1997 blockbuster movie. "I started with a few notes," Cameron recalled, "and threw in a lot of design and other ideas from earlier scripts or story fragments — things I'd written as far back as the '70s. Its narrative roots lay in my childhood enjoyment of stories like Edgar Rice Burroughs' John Carter of Mars or Rudyard Kipling's The Man Who Would Be King — the classic idea of a Westerner, a military guy, plunked down into a different culture and having to earn his way into that culture."

As Cameron originally envisioned the project, he would film live-action plates for Avatar in rainforests, and then add the CG Na'vi and avatar characters and creatures to those plates, just as Stephen Spielberg had populated his Jurassic Park, in part, with digital dinosaurs. But, performance-wise, there was a huge chasm separating an ambling, expressionless CG brachiosaur, on screen for a few minutes, and emotive, humanoid characters that would have to carry a feature-length film. What Cameron needed to make Avatar a reality was a way to motion capture not just the gross body movement of an actor, but every nuance of that actor's facial performance.

"Our goal in using performance capture," noted Avatar producer and longtime Cameron collaborator Jon Landau, "was not to replace the actor with our computer animated character, but to preserve the actor — because what a great actor does and what a great animator does are antithetical to each other. A great actor withholds information. Dustin Hoffman can sit there and do nothing in All the President's Men, and you are riveted. But no animator would ever animate a character to just sit there and do nothing. We wanted performance capture to be the 21st-century version of prosthetics, something that would allow actors to play fantastic characters that they could not otherwise play."

Fourteen years ago, filmmaker James Cameron wrote a treatment for Avatar, an ambitious propel intended to devote computer generated effects to unprecedented levels of artistry and sophistication. Set on a mineral-rich moon called Pandora, the story centers on an avatar program that combines human DNA with that of the indigenous people, the Na'vi, to create a hybrid creature capable of breathing Pandoras toxic atmosphere and thus able to work in the mines of an exploitive earth-based consortium. Conscripted into the avatar program is Jake Sully (Sam Worthington), a paraplegic combat vet who comes in love the Na'vi and lead them in their existential fight against the human intruders. From the project's inception, Cameron intended to realize the Na'vi and avatars as CG characters that would be brought to 6fe through the capture of actor performances. Though performance capture was still in Its embryonic stages at the time Cameron conceived Avatar, the technology finally caught up to his vision in 2005, at which time the director and his technological team began conducting tests and building hardware for a new filmmaking paradigm they called virtual production. Cameron directs odors and stunt people on the performance capture stage at Playa Vista, where the virtual production was housed.

Jake and his fellow recruits journey to Pandora aboard the Venture Star, a massive interstellar spaceship measuring a half-mile long. The space shots were created by Blur, one of several effects vendors that provided postproduction support for the main body of work being realized by Weta Digital. The all-CG space sequence featured a Venture Star model built in XSI by the production's in-house art department, which was then modified and re-textured by Blur to be used in its 3ds Max pipeline.

Gross body motion capture had become a relatively routine filmmaking technique by the time Cameron and Landau began considering the Avatar project; and, in fact, Cameron had used motion capture to create digital background passengers aboard Titanic. "Even back on Titanic," commented Cameron, "we knew how to put big marker balls on people and get a skeleton from motion capture. But we didn't know how to capture faces. Faces were still hand-animated. The thing that had not been cracked, the thing that would have to be cracked to do this movie, was real human facial response in CG. Nobody had done that yet." A thorough examination of existing facial capture technology led Cameron and Landau to the conclusion that it was still in its embryonic stages, and that Avatar's time had not yet come. "I went off and did Titanic, figuring I would look at Avatar again afterward. There was no rush to make it because it was a timeless project. I knew it would still be a good idea in two years — or 20 years — and so I was content to stick it in a drawer for a while."

Avatar gathered dust in that drawer for years after the completion of Titanic, as Cameron became interested in 3D and documentary filmmaking as means for exploring space and underwater realms. In the course of planning a Mars film, he and Vince Pace, a leading developer of digital camera systems, engineered a stereoscopic camera rig that they subsequently took out on a series of ocean expeditions. The results were Ghosts of the Abyss and Aliens of the Deep, stereoscopic films that documented the Titanic wreckage and deep-sea life forms, respectively.

When Cameron turned his attention back to feature filmmaking in early 2005. he considered as his next stereo 3D project either Avatar or Battle Angel, another script he and a co-writer had been working on for years. "Battle Angel just wasn't gelling at the time," said Cameron, "and so I dug out Avatar and read it at one sitting — and it was much better than I remembered it. So I decided to put it in motion, and I pitched it at Fox." Cameron asked the studio for a significant sum of money, which he would use to prove — or disprove — Avatar's viability. "I told them: 'We'll look into the capture technology. We'll figure out who our vendors are going to be. We'll develop our pipeline. And parallel with that, we'll do the creature and character design. After you've seen all the production design, and after you've seen that the technology will work, if you still want to bail out, you can bail out, no hard feelings.' By May 2005, we were putting together an art team and a tech team, with Rob Legato spearheading the tech development."

The Venture Star nears the planet Polyphemus and its largest moon, Pandora. Guided initially by artwork provided by production and Weta Digital, Blur created the shot by projecting matte paintings onto 3D geometry. The look of the planet and moon continued lo evolve throughout production. Posed on input from Cameron. Due to the film's intended stereo 3D and Imax-format presentation, every one of its 2,500 effects shots had to be rendered in stereo and detailed to hold up at 4K resolution.

Legato, Cameron's friend and visual effects supervisor on Titanic, had recently come to him with an intriguing concept he had developed while working on the CG crash sequence in Martin Scorsese's The Aviator. As visual effects supervisor on that film. Legato had been frustrated by the detached nature of directing CG shots: describing camera moves to CG artists; waiting days or weeks while they produced a first attempt based on that description; reviewing the results, finding them not exactly what he had in mind; then repeating the entire process again — and again, and again. "People who are camera-oriented," noted Legato, "find it difficult to explain what they want and have other people interpret that in the computer. When we're on a set, we do 20 iterations of a shot in a minute — a little higher, a little wider, a little tighter, a little to the right. Within a minute, you've done what you would have tried to explain to CG people 14, 15 different times, with all the reviews in between. Everybody has their own camera eye and sensibility, and to translate that through an intermediary is very difficult. You get frustrated, and find yourself wanting to say, 'Just give me the damn camera and let me show you what I mean!"

Essentially, Legato had done just that for The Aviator — grabbed the camera from the CG artists so he could show them what he meant. Inspired by the realtime joystick control capabilities of his son's videogames. Legato and his team of previz artists had used off-the-shelf software and hardware to jury-rig a system that would enable him to move a virtual camera through a CG scene. "We came up with something that would accept live input," said Legato, "which in this case was pan and tilt wheels. I'd pick an animation of the plane and then shoot it, panning and tilting, trying a 14mm lens, then an 18mm lenses, experimenting with this and that. Ideas came to me as I was shooting, just as they do when I'm shooting live-action. I shot everything long, from several different vantage points; and then I cut it on the Avid. And when I cut it, 1 found there was the same energy you get when you're cutting live-action material. It wasn't like it typically is in CG, where you tell somebody to do an 8.6-second shot, and they give you 8.6 seconds, and no more. You don't get to choose between five angles of the same action, shot from five different cameras. You don't get the wobbles and mistakes you never intended to be there, but that end up helping the shot. Being able to move the camera in CO in real time gave all that back to me, and the camera work was mine. CG tends to neuter your distinct personality as a filmmaker, because it has to go through the filter of someone else — the CG artist. Once you remove that filter, it is all you again."

Cameron shot most of Avatar's live-action on Weta soundstages in New Zealand, a decision he and producer Jon Landau made upon visiting the facility early in the project's development phase. Production designer Rick Carter built a number of full interior sets for the live-action shoot, but many live-action setups invoked only small part'd sets on greenscreen stages, extended with expansive CG environments built and tracked into scenes by Weta Digital and other vendors. Sam Worthington was filmed on a virtually empty greenscreen stage for a shot of wheelchair-bound Jake exiting the landed Valkyrie shuttle at Hell's Gate, the humans' base of military, industrial and scientific operations on Pandora. The Hell's Gate shot was one of 70 contributed by Framestore, which built an array of 3D structures — including the base's headquarters, refinery, perimeter fence and guard towers — and then brought the environment to life with animated aircraft and vehicles, a mix of digital assets built at framestore, Weta Digital and Industrial Light & Magic. Framestore completed the shot by adding atmospheric effects and digital doubles in the background.

Knowing that the camera-oriented Cameron would appreciate the concept. Legato gave him a call. "I explained the idea to Jim, and he got intrigued by it," Legato recalled, "and he realized that it would be ideal for Avatar." Cameron instantly recognized that the virtual production setup would enable him to create and shoot not just CG characters and creatures for Avatar, but also CG environments, eliminating the need to constrain himself to live-action plates shot in rainforests. "The idea was to come up with something that would enable him to walk out onto the motion capture floor with a camera that felt familiar to him. look through an eyepiece or monitor, and see computer generated characters in a computer generated set."

"With this virtual camera," Jon Landau elaborated, "when Jim pointed the camera at Sigourney Weaver and looked into the camera, he wouldn't see Sigourney Weaver — he would see a 1990s videogame version of Weaver's avatar character, in real time. And as he moved the virtual camera around the stage, he wouldn't see the stage — he'd see the world that Sigourney Weaver's character was supposed to be in. So Jim could be out there operating the camera, working as intimately with the cast as he would with Kate Winslet or Leonardo DiCaprio, but seeing what his shots would be in CGI."

To oversee the technical development of virtual production, Cameron and Landau hired Glenn Derry who, with his Technoprops hardware company, had realized some of the technical requirements for Rob Legato's Aviator project. Derry teamed with Motion Analysis Corporation, which worked with Lightstorm during the protracted R&D phase, applying its existing real-time motion capture capabilities to the task of devising a production pipeline protocol for multiple-character capture. Derry also brought on previz artist Nolan Murtha, who would become the show's digital effects supervisor for virtual production.

Simultaneous to the testing of technology, a core group of artists — Wayne Barlowe, Yuri Bartoli, Neville Page and Jordu Schell — began producing concept illustrations of the Na'vi, avatars and Pandora 11 creatures, initially working at the kitchen table at 'lower house,' a structure neighboring Cameron's home in Malibu that he had purchased to use as a workspace. "It's kind of my man cave," Cameron said. "No furniture — just a bunch of art tables. And I stuck this core group of creature designers out there."

When he was in town, Cameron would often join the artists at the art tables to do his own sketches. On some days, he would bring in artwork he had rendered for his previous films as inspiration for the team. "That really kept us on our toes," said Yuri Bartoli, "because it showed us that Jim had the same skills we had. He didn't want to see finished paintings from us at this point. He didn't want us wasting time rendering something and making it look too good. It was all pencil-and-paper drawing, just sketches of the Na'viand the creatures, mainly, so he could see what direction we were headed in."

Cameron had already rendered fairly detailed sketches of the Na'vi, illustrating a people with large, cat-like eyes, broad leonine noses, large articulating ears and tall, lean, muscular bodies. Using Cameron's sketches as a springboard, the designers initially explored less anthropomorphic Na'vi designs. "We had everything from a Cyclops version to a multi-eyed version of the Na'vi atone point," Jon Landau recounted, "but eventually we went back to a much more traditional foundation. We came to realize that if we wanted the audience to relate emotionally to these characters, there needed to be familiar touchstones. So they're blue, and their eyes are slightly wider and their mouths are smaller and their noses are broader — but they still have what is basically a human face."

The concept artists and other designers brought onto the show would refine Na'vi features for another two years; but, throughout, Cameron never wavered in the decision that they would have blue flesh. "I visualized the Na'vi as blue right from the get-go," Cameron remarked. "It may have been as simple a thing as, 'Well, green's already been taken' — which it was. Everything from the Wicked Witch of the West to every Martian ever created to the Hulk was green. So green was taken, but blue wasn't — at least not then — and 1 thought blue-skinned characters would be cool."

The avatars, due to the DMA of their human operators, would sport a lighter shade of blue and slightly more human-looking facial features. The concept artists generated hundreds of sketches to determine the appropriate mix of human and Na'vi characteristics, illustrating avatars with rounder, smaller and more human ears, for example, and avatars with very large, animalistic ears — and everything in between. Further down the road, when the roles had been cast, the designers would render avatar looks that incorporated the specific features of the actors portraying them so that the Jake avatar would resemble Sam Worthington, Grace's avatar would resemble Sigourney Weaver, and the avatar for Norm, a scientist who arrives on Pandora with Jake, would resemble actor Joel David Moore; but at this early stage, the artists simply combined Na'vi features with those of random actors and models. "We would take generic photos of people and develop the look from those," said Yuri Bartoli. "We stayed away from famous actors completely, because Jim didn't want to get used to seeing an avatar that looked like a particular, known actor."

Jake stares in wonder at his avatar, floating in a birthing chamber in the Hell's Gate compound Rick Carter's set included two tubular acrylic chambers bulk by Richard Taylor's Weta Workshop, a project that took two years to complete due to the engineering requirements of the units' complex, slightly futuristic mechanisms. Weta Workshop came onto the show during preproduction to design and build much of the requisite military and mining hardware, as well as costuming, jewelry, weaponry and other cultural accouterments associated with the Na'vi.

As the concept artists moved from lower house to offices at Lightstorm Entertainment — Cameron's production company in Santa Monica — they continued to design characters, as well as Pandora's array of creatures, bringing approved sketches into Photoshop for color painting and detail. Once a painted rendering was approved, the artists would take it into ZBrush, where they transformed the 2D concept into a 3D digital sculpt. "Working digitally, we were able to do many different versions quickly — longer horns, shorter horns, bigger eyes, smaller eyes," Bartoli recalled. "It was very interactive, and Jim could sit there and drive the whole process. Jordu Schell was also doing a lot of physical clay sculptures, particularly of the Na'vi."

One of the early creature concepts was that Pandora's terrestrial animals would have six limbs; but Cameron, not prone to embracing fantasy elements for their own sake, engaged the designers in a series of discussions regarding the potential scientific basis for the multi-dexterous creatures before fully approving the idea. "We were always working from a biological structure," stated Bartoli, "rather than making things up out of thin air. So we gave a lot of thought as to why these creatures might have six legs, when four is more efficient from an evolutionary standpoint. The answer was that Pandora has very uneven terrain, with a lot of vines and overgrown jungle; and so they use these extra limbs to grip onto things and pull themselves up. Instead of paws, the extra limbs have what look like hands on them, enabling the creatures to grab their prey, and still have four limbs firmly on the ground."

To determine if the extra limbs would look appropriately graceful and powerful in motion, the team brought in ILM animator Paul Kavanagh to animate a series of six-legged walk and run cycles. Kavanagh then produced a successful animation test that featured a viperwolf stalking prey, and then accelerating into a gallop. "When I saw that test," said Cameron, "I thought: That's it. We're on. All of the terrestrial creatures will be six-legged."

Superviser John Rosengrant and his team at Stan Winston Studio — now legacy Effects — also contributed a number of designs and practical props to the show. To aid in lighting and establish eyelines on the live-action set, SWS created generic mole and female Na'vi models at fail scale, applying intricate paint schemes designed in Photoshop. The Winston team also built human-scaled busts of the main Na'vi and avatar characters, incorporating the performance capture actors' facial features to ensure they would be recognizable even in their nonhuman forms. SWS hair department artist Aimee Macabeo works on the life-sized Na'vi female.

Among the six-limbed creatures is the thanator that chases Jake through the jungle early in the film, and also plays a pivotal role in the final battle. Though several of the artists took a crack at the thanator, the final design had to be credited mostly to Jim Cameron, who rendered a number of sketches of the large cat-like creature. Concept artists were more directly responsible for the design of the hammerhead titanothere, a rhinoceros-like beast with anvil-shaped horns. After the Photoshop rendering and ZBrush sculpt were approved, designer Daphne Yap — a new member of the expanded art department — painted intricate and colorful markings that would cover the hammerhead's body. Yap created similarly detailed marking patterns for all of the creatures, a step the filmmakers dubbed 'Daphnization.'

The extra limbs on the terrestrial beasts led to the idea of an extra pair of wings on the flying creatures such as the banshee and the leonopteryx. The former would be featured in several key scenes — including one in which Jake claims a banshee as his own for an aerial hunt and, later, a romantic flight scene — while the leonopteryx would play a crucial role in the final battle sequence. Wayne Barlowe, Neville Page and Daphne Yap rendered banshee concepts, all aimed toward finding a new take on the ubiquitous flying reptile. "At first glance," stated Cameron, "the banshees are somewhat like a dragon or a pterodactyl. But we weren't interested in doing just another dragon or pterodactyl. We wanted to do something different. The final banshee has some aspects of bats, some aspects of eagles; but when you really study them, you can see that they are very different from what we've seen before." A unique feature of the banshee design was the structure of its mouth, which had a two-part upper section that would pull back as rows of teeth moved forward. Yuri Bartoli sculpted the banshee jaw in both an open and closed position in ZBrush to illustrate the articulation. Later, the jaw action was engineered and further refined in practical models built by Stan Winston Studio.

As the months passed, Yuri Bartoli continued to tweak the banshee design, painting detailed textures, colors and markings. "I started doing some very high-resolution treatments in Photoshop," stated Bartoli. "painting megapixel images so that every tiny pore and marking was in these paintings. Then I used Z-Brush to project the textures onto a banshee model to see how it would look in 3D. I did the same thing for the leonopteryx. For all these creatures, we worked at a tremendous level of detail. At one point, I found myself working on the detail of the molars in the back of the mouth of the hexapede — a kind of gazelle that makes a brief appearance —and I thought: 'Is anybody really going to see this? Am I going too far?"

Cameron directs Sam Worthington and Sigourney Weaver — playing scientist Grace Augustine — on the performance capture stage. Rather thon capture facial performance separately within a head volume equipped with multiple closeup cameras, Cameron proposed emulating a live-action shoot by capturing the facial and body performances simultaneously, as actors interacted with one another on stage. The live-action-style shoot was made possible, in part, by food) capture headgear and by the ability of Giant Studios' motion capture system to 'fill in' areas of the body occluded by set pieces and multiple odors within the volume.

The live-action paradigm was also facilitated by the development of a virtual camera — essentially a monitor linked to a computer system that could stream captured performance and camera movement in real time and map them to primitive videogame versions of CG characters and environments. The system enabled Cameron to operate his camera in the virtual world just as he would a traditional motion picture camera In the real one. As built by virtual production supervisor Glenn Derry and his crew at Technoprops, the camera featured a series of hand controls that enabled the operator to simulate zoom, dolly, crone, Steadicam and other moves.

The technical endeavor guided by Rob Legato advanced apace with the designers throughout that summer of 2005; and by the time Cameron returned from a two-month Titanic expedition at the end of August, Legato and his small crew had set up a temporary pipeline that enabled them to feed live mocap data into software that created an instant, realtime visualization of a digital character enacted by a performance-captured subject. They really had the system on its feet," recalled Cameron. "In fact, it took me pretty much a whole day to understand what they were doing because they'd taken it far beyond what I'd originally imagined." The technical team had) focused its efforts on facial capture, exploring each of the two schools of thought that currently prevailed. One approach dictated gathering extremely detailed 3D data by building a mocap volume specifically for the head and capturing facial performance from multiple close-range cameras, using software that could track markers and triangulate points in space. Such an approach would result in a huge amount of 3D detail that would enable effects artists to reconstruct a nearly perfect facial mesh.

But capturing the face separately from the body, and in distinct sessions that would take place after Cameron had directed the actors together on the mocap stage, violated the live-action paradigm Cameron and Landau had envisioned for Avatar. On top of that, the precise data captured from the actor's face would not have been particularly useful in animating his CG counterpart. "We were going to be retargeting the facial capture onto a Na'vi or avatar face, anyway," explained Glenn Derry, "which wasn't the same face shape as the actor's. This wasn't Brad Pitt playing Brad Pitt as Benjamin Button. This was the very human-looking Zoe Saldana playing the Na'vi Neytiri. Since there was such a difference in the faces of Zoe and Neytiri, we were never going to get a one-to-one correlation between the mocap data and the character's face, anyway. Even in the best retargeting, we would still have to get an artist in there to work with it and make it perfect."

The second school of thought on the subject favored an image-based approach, using optical flow technology to capture a lower level of facial detail from cameras as the actor performed, in conjunction with the body motion capture, and then expending more resources and time at the back end to build extremely robust, sophisticated facial rigs that could extrapolate from the lower-end stage data. As far back as 2002, a Lightstorm team had tested image-based performance capture for a project called Brother Termite. Shooting a scene at a cemetery, the crew had turned the footage over to a visual effects company, which then produced a finished sequence. "That showed us that image-based performance capture was better, even back then," Jon Landau noted. Cameron proposed doing image-based facial capture for Avatar using a single camera placed directly in front of the actor's face; but there was resistance to the idea, initially, with some insisting that more detailed data would have to be captured to create a convincing facial performance in CG. "Early on," said Cameron, "people thought we needed more cameras, that we would have to photograph the entire face. But I kept saying: 'We don't need to see the entire face. What we need is a camera that can see what the eyes and the mouth are doing; and then, we need a model that is so thoroughly and well rigged, that when the camera image shows that the corner of the mouth is pulling back, the system knows what to do from there.' If we rigged that model properly, when the mouth pulled back, the system would be able to figure out what else was happening — even though you couldn't see it in the single-camera reference. All we needed from the stage was the core data of what the actor was doing."

Glenn Derry supported Cameron's approach. "A single camera in front of the face at standard definition would give us a lot more information than a high-definition camera from across the stage," Derry noted. "It would also make our facial capture system mobile, so it would work while the actor was performing, without having to separate the face from the body. The capture would be easier to acquire, faster to track and process, and it would get us to a 90-percent threshold much more quickly. And if worse came to worse and we didn't get what we needed, we could still hand-animate the face, because we would have bullet-proof reference of every single frame from this camera that was right in front of the actor's face the whole time."

Glenn Derry and his crew also built the head gear for for facial capture, mounting a small standard-definition NTSE camera to the end of a six-inch boom arm that was attached to a fiberglass and foam helmet custom-built from Stan Winston Studio lifecasts of the actors. Much like a head-mike worn by concert performers, the boom arm maintained the position of the camera directly in front of the actor's face, regardless of his head or body movement. Photographed by multiple HD witness cameras for animation reference. Sam Worthington, as avatar Jake, plays a performance capture scene with Zoe Saldana, as the Na'vi princess, Neytiri. Ordered to infiltrate the Na'vi, Jake falls in love with both their culture and their princess. After performance capture sessions with the actors — recorded by 120 mocap cameras mounted to the ceiling and multiple HD cameras on stage Cameron added final camera to the low-rez scene, creating a template that was subsequently delivered to Weta Digital and visual effects supervisor Joe Letteri. Template files included alphanumeric codes generated by Gaia, the productions digital asset management system, which tracked performance capture takes, camera takes and successive iterations of shots. The system also tracked stereo-related data such as focal lengths, interocular and convergence distances.

At Weta Digital, the layout department replaced template elements with higher-resolution assets, then passed the shot on to the animation department, headed by animation director Andy Jones. Though performance capture technology got character performances 90 percent of the way to final, every shot required the artful eye and hand of an animator to deliver the final 10 percent. The single-camera head rig supplied raw facial capture data, but was of little use as animation reference because the proximity of the camera to the actor's face resulted in a fish-eye-lens look. Animators relied heavily on HD witness camera footage to recreate the actors' original performances. The photorealism of Weta Digital's final shots was the product of a new, more physically based pipeline built from scratch over a yearlong research and development phase. As part of the physical approach, Weta built facial models for main characters such as Jake and Neytiri with anatomically correct muscles, fat and tissue. Weta also upgraded its approach to lighting, moving from ambient occlusion to the more streamlined — but ultimately more realistic — spherical harmonics, which determines light-source direction for every given point on a surface.

When the issue of the single video camera had been settled, Cameron worked out specs and engineering details with Glenn Derry, who would build the system. "I wanted the smallest possible camera," said Cameron, "with the widest possible lens, sitting at the end of a very lightweight, thin, six-inch boom. And I wanted all of that attached to a skullcap that would conform to the actor's head. It wasn't like we could put neurosurgical screws into the actor's skull to support the camera. I proposed that idea, but the actors said no. I said:'Come on! Where's your dedication to your art?' But they wouldn't go for it. So I figured we'd make this skullcap, getting as much point of contact with the head as we could so that when they moved, the camera would remain stable. The less movement and instability there was in the camera, the better the dataset we'd get out of it."

Not only had the tech team worked out a basic methodology for facial capture by the time Cameron returned from his Titanic expedition, Derry and his crew at Technoprops had also built a beta version of the virtual camera. Assuming a director would be most comfortable with a camera that seemed familiar, Derry had built the virtual camera so that it would look and handle much like a typical motion picture camera, complete with tubular eyepiece. But the virtual camera was not really a camera at all, but rather a somewhat camera-shaped object with an interface that could stream motion capture in real time and map it to digital characters within digital environments. "The real 'camera' was in the interface that tied directly into MotionBuilder," Derry explained. "That interface would give Jim the ability to scale things, to fly around, to do everything a camera operator would do."

The controls enabled the camera operator to replicate any number of camera moves, such as zoom, dolly and Steadicam, and also allowed for scale variations. "Depending on how you set the scale," noted Cameron, "the camera could move through an environment in a one-to-one relationship, or it could move through it as if it were a tabletop miniature. I'd be holding the camera, and they could set the scale so that suddenly I was 60 feet tall, moving the camera around this miniature environment." Scaling enabled Cameron to simulate large crane moves on stage, merely by swooping his arms with the camera in hand. Or he could set the parameters of a crane move — starting at, say, 100 feet in height and ending in a closeup as a character said a specific line — and the computer would build that curve.

The virtual camera system also allowed for speed variables. If a creature was running through the jungle at high speed, for example, traveling hundreds of feet, the virtual camera operator could easily keep up with the CG creature through platforming. "We would platform to a path, and move through space as if on a dolly," said Cameron, "so all we had to do was follow the subject. We could platform to anything — a camera path, a character, an object, a vehicle. It was one of the really fundamental ideas. This was the kind of stuff the tech team came up with initially and presented back to me — and I was dazzled by it."

With the prototype facial capture head rig, realtime motion capture and virtual camera all at some level of functionality, Cameron decided to create a proof-of-concept test, shooting the scene from his treatment in which Jake first meets Neytiri in the forest. "I wrote that out to a five-page scene," recalled Cameron, "and then we brought in a couple of actors to shoot the test. I saw immediately that they were not in any way daunted by the process, that they were not particularly concerned about having this helmet stuck on their heads. For them, it was blackbox theater. It was pure acting. It was all about the emotion and reacting to each other — no hitting marks, no worrying about lighting, costumes, makeup and hair. It worked well."

Cameron and his crew shot three minutes of material, with both motion capture and HD cameras, and then sent the mocap data to Industrial Light 6c Magic where, under the supervision of Dennis Muren, a team fully animated and rendered a 35-second scene. Cameron and Landau screened the test for executives at Twentieth Century Fox in a split-screen presentation that showed HD video reference of the original actor facial performance alongside the final rendered animation. "Jim was showing them that you could extract an emotive performance using this technology," said Nolan Murtha. "The character performances were absolutely believable, and almost exactly matched the actors' performances." "This prototype scene was a good representation of where the project could go," added animation director Richard Baneham, who initially joined the Avatar project to oversee the test animation and then stayed on for four years. "It gave us a broad understanding of the potential of the technology. It seemed that it could give us a digital performance that was sustained and could connect with an audience. That was the glimmer of hope born in the prototype."

Of all the Avatar innovations, the one with the greatest potential for widespread application to the film industry was 'Simulcam,' which fed CG imagery to the live-action camera. executing real-time low-rez comps that enabled Cameron to see in his viewfinder and on monitors digital set extentions rather than greenscreen, and digital characters interacting with live environments and actors. Though Cameron made ubiquitous use of Simulcam throughout the live-action shoot, it proved most advantageous in shooting the scene in which Jake first awakens in his avatar body in the science lab.

Although Fox would not officially greenlight the production for another year, studio executives and Cameron came away from the test convinced that technology had finally caught up to Avatar. Eighteen months out from the start of production, Glenn Derry and his team began building the technological infrastructure to support the project. "I launched into the toy-making at that point." recalled Derry, "building the actual virtual camera and the facial capture system we would use to make the film, and going into physical production of the mechanical and electronic components of the head rigs. We made the cameras — which were just standard-definition NTSE cameras — and we made the recording systems, the electronics that drove everything, the transmission system that would send the video back to us." Camera and booms were ultimately attached to fiberglass and soft-foam helmets fashioned from Stan Winston Studio lifecasts of actors portraying the Na'vi and avatar characters. Some helmets had long strands of material attached, to help the actors remain aware of the Na'vi and avatar braids that would adorn their digital characters. Judging the initial version as too cumbersome. Derry took another pass at the virtual camera at this point, as well, building what he called the 'ring cam.' "That one looked like a steering wheel type of camera system," said Derry. "But even that was unnecessarily awkward. Ultimately, when we boiled it down to what we would be using of this 'camera' we realized all we really needed was a view port of some sort — a monitor — and the controls. That led to the 'swing cam,' which is the camera Jim ended up using on Avatar. It had a mechanism so that the monitor would always be facing him, no matter how he moved the camera." The swing cam weighed seven pounds, light enough for Cameron to manipulate easily and carry for long periods of time on stage. "Later, we built versions that weighed only two pounds; but by then, Jim had become used to the feel of the heavier one, and that's the one he continued to use."

The hardware build would go hand-in-hand with the development of software and a new production pipeline by the artists and technicians at Weta Digital, whom Cameron and Landau had approached about Avatar at the inception of the project. The filmmakers and Weta Digital visual effects supervisor Joe Letteri engaged in more pointed talks in early 2006, discussing techniques the company had used for Peter Jackson's King Kong. Among the most critical topics of conversation was facial capture. "We discussed how we were going to move motion capture away from a postproduction process, which is what it had been in the past," recalled Letteri. "For Gollum, we had shot all the plates; and then we captured Gollum's performance with Andy Serkis. But Jim wanted to do all that up front, to have it be the equivalent of an onstage setup. We had also used track markers on Andy Serkis' face, but Jim did not want to go that route, either. He thought it would be too restrictive to deal with that every day. Instead, he wanted to go with a head-mounted system. So we began pursuing that, talking with Glenn Derry about how we would build the pipeline."

Cameron first shot performance capture of Sam Worthington acting the scene on a bare-bones lab set that was scaled down to accommodate the difference between Worthington's size and that of Jake's avatar. / On the full-scale set, Cameron blocked out the scene with the supporting players, using one of the Stan Winston Studio avatar models to determine lighting and give the actors a sense of the CG character's relative size in the space.

Simulcam required a mobile motion capture system that could be moved easily from set to set and could function. In lighting environments very unlike the sterile, flat-lit volumes most amenable to infrared mocap cameras. Glenn Deny devised an active light emitter that eliminated interference from set fighting by essentially canceling it out. An extendable rod positioned at the height of Jake's avatar helped actors maintain correct eyelines. / Sam Worthington's captured performance from earlier in the day was mapped onto the CG avatar, allowing Cameron to frame and follow the character with a handheld live-action camera as he shot the scene.

As a demonstration of an on-stage performance capture setup. Letteri and a small crew motion captured performer Shane Rangi in one of Derry's head rigs — with small camera attached — and used the streaming data to drive a CG Andy Serkis face the team happened to have rigged for another project. "We did this as a demo for Jim," stated Letteri. "We knew he was coming down to talk to us about the project, and we thought. 'Wouldn't it be great if he could see a digital character performing live while he was on stage — not just the body, but the face, too?' When Jim came down, we showed him this proof-of-concept, and we all agreed that it was a good start. Once we had that, we knew we'd be able to improve the process."

"At that point," said Derry, "Weta came on board, and said, 'Okay, you handle the physical aspects of this, the hardware, and we'll handle the software end,' That whole time was great for all of us, because nobody was telling us we couldn't do it. Jim was really pushing us, but we had time to figure it all out — which was a once-in-a-lifetime experience because Hollywood, in general, doesn't work like that. You don't get to R&D like that. Usually, everything in Hollywood is based on what has been done before, on this movie or that movie. Well, we didn't care how things had been done before. We were moving forward."

The design team, too, was moving forward, doing more detailed work on creatures and characters and initiating designs of vehicles, spacecraft and other hardware — a task that fell primarily to concept artists TyRuben Ellingson and Ben Procter, both of whom had joined the art department specifically for that purpose. After the initial design phase, Ellingson worked with Cameron to generate highly detailed 3D hardware models. "These design models were so detailed," Ellingson observed, "they often went into the pipeline. They went way beyond the cosmetic illustrations most often associated with conceptual design, and they allowed for motion tests to be done very early in the production."

A prominently featured piece of military hardware was the 'Amplified Mobility Platform,' or Ampsuit, a kind of walking tank with a human driver at the controls and an arsenal of high-powered weapons. The Ampsuit design models were ultimately sent to Stan Winston Studio, which would build a practical, full-scale Ampsuit for production, and to Weta Digital, which would build ambulatory CG versions. "Jim is, by nature, deeply interested in details pertaining to how things work," noted Ellingson, "and so he and I had already designed and trouble-shot the complex shoulder and limb jointing of the suit before the model was shared with Weta or Stan Winston's. This kind of 'deep thinking' up front, before the visual effects companies got onboard, allowed Jim to move forward with a very high, verifiable confidence in what he wanted these machines to look like and do. It was a very rich way of working, a very rewarding and potent process to be part of."

Human consciousness is transferred to the body of the avatar through a 'psionic link unit' built as a practical prop by Weta Workshop. Buf Compagnie designed and executed the vertiginous consciousness-transfer effect in 3D, layering multiple CG tunnels of various diameters and textures, some rendered to resemble networks of neurons, others rendered to appear as rays of light or lightning flashes. Buf generated each tunnel independently, according to its trajectory from camera, and then composited it with the others to achieve a sense of depth and volume.

The designers also began to focus on Pandora's exotic environments, realized under the stewardship of Robert Stromberg, whose contributions to the film's design would be so significant, he would earn a production designer credit, along with Rick Carter. "Even though the Pandora environments would be virtual," Stromberg commented, "they went through the same growing pains you'd have with any set design, and we had to follow all the guidelines of traditional production design." Stromberg expanded the Lightstorm design team, and also created a virtual art department that would build digital assets for the shoot, overseeing their development from design to final product. At the same time, Yuri Bartoli segued from conceptual work to the virtual environments team, and was responsible for building many of the assets. The design and virtual art departments worked side by side at Lightstorm initially; and then, later, the virtual art department moved to the stages at Playa Vista — the former Hughes Aircraft facility, now a filmmaking complex — where the performance capture would take place.

Stromberg's initial concern was the overall, bird's-eye view of Pandora; and then, as the design effort progressed over the year, the artists zeroed in on specific settings and plant life. "We started to dissect Pandora into smaller and smaller pieces," said Stromberg, "until, eventually, we were designing individual plants." In addition to creating paintings to illustrate the environments in general, the art team rendered keyframes of scenes from the movie, as described in Cameron's treatment. To suggest the glowing bio-luminescence of Pandoran plant life, as seen at night, the artists drew scenes in neon-colored pencil on black paper, a technique Cameron had used to illustrate the look of the glowing aliens and their craft in The Abyss.

In addition to the bioluminescence, a visual motif repeated throughout Pandora's environments, characters and creatures, Cameron had imagined the entire rainforest as having a cyan tint, in contrast to earth's green. "That sounded like a good idea," said Stromberg, "but we started to realize that shades of blue are what we use to create atmospheric depth; and so. if everything was that color, it could be a problem. Over time, we started bringing more greens into the environments. That not only helped us with creating depth, but it also gave the audience a psychological attachment to this alien place, because it looked more familiar. The rainforest evolved from something that was much more alien to something that was a bit more traditional."

The interior of Home Tree, the massive tree in which Neytiri's tribe lives, also evolved considerably, starting as a five-level structure and ending with only two. "We realized that when you were inside it," said Stromberg, "all those levels gave it a boxed-in feeling. So we opened it up by giving it just a ground floor and an upper deck. It made for more dramatic angles looking up into the tree from inside. We also had to work a lot of physics into the design of this enormous tree, giving it a root and base structure that would actually support a tree of that size." Another iconic tree is that at the center of the Tree of Souls,' the sacred site that embodies the spirits of Na'vi ancestors. "That tree had to have a lot of personality, but a very different personality than the Home Tree. It had to have a much more ancient look. We looked at bonsai trees and got a feel for a kind of classic knotted and twisted look — which, again, helped the audience connect to the tree because it was a shape they recognized and were comfortable with. The tree also had glowing, willow-like tendrils."

With Cameron's approval of the designs, the virtual art department began building proxy assets — low-polygonal-count representations of every fantastic plant, tree, creature and character on Pandora. Motion capture would be streamed to those assets in real time during the shoot, creating low-rez animation and environments displayed on the virtual camera and surrounding monitors. "Throughout 2006, we were essentially building the 'stuff we needed to make the movie," said Nolan Murtha. "We built characters, environments, hardware, plants, always working closely with the concept and sculpture artists to make sure our assets would match the original designs. Everything needed for production was built in Maya for Motion Builder by this in-house team."

After the digital character assets had been built, Richard Baneham and his team began developing their animation rigs, exploring creature and character movement and behavior. "W^re worked with the skeletal structures to see how they would function from a practical standpoint," said Baneham, "and how they would move. We explored the animation of everything from the smallest creatures — like the insects we see moving around on plants — all the way up to the leonopteryx and the thanator."

Once Weta Digital had gone into shot production, its artists also built puppet versions of their high-rez creature and character models for playback on the mocap stage. "By building these puppet versions," noted Joe Letteri, "all the skeletons and rigging were correct to what the final models would be; and the puppets looked like what the final models would be, in general."

Meanwhile, Giant Studios was setting up the motion capture system on stage at Playa Vista — coincidentally, in Hie building next door to Giant's location at the time — mounting 120 infrared cameras onto a ceiling grid. Giant Studios had won the performance capture assignment for Avatar after a lengthy takeoff process, primarily due to its advanced real-time capabilities and highly intuitive biomechanical solve. "If their system missed seeing a marker or two or three." said Jon Landau, "it still could interpret, based on the other markers, where the skeleton for the gross body needed to be. With the other systems we tested, we got into a little bit of the 'watusi,' where the arm would be in one position, but the system would think the arm was somewhere else."

The ability of Giant's system to extrapolate and fill in occluded areas was key. as the filmmakers planned to shoot the actors with multiple high-definition video cameras inside the mocap volume — one acting more or less as the 'A' camera and the others recording additional crucial reference. "The need to see every marker at every point in time would have severely limited where we could place our virtual camera and reference cameras," observed Landau. "It also would have limited where we could place the actors in relation to one another. Because Giant Studios' system could extrapolate what the movement would be, even if part of a body were occluded, we wouldn't have to worry as much about blocking the performance capture cameras. We could bring people and HD cameras into the volume. We could have set pieces on the stage. Actors could stand close together or wrestle on the ground."

Jake converses with Colonel Quaritch (Stephen Lang) at the controls of an Amplified Mobility Platform or 'Ampsuit' Envisioned as a human-piloted walking tank with high-powered weapons, the Ampsuit was designed by concept artist TyRuben Ellingson with much input from Cameron. Production sent designs and a crude CG model employed for motion studies to Stan Winston Studio, charged with building a full-scale Ampsuit for use in live-action situations. / Framestore piled out the original greenscreen plate with a 3D armor bay — complete with animated digital doubles and vehicles — and incorporated parked Weta Digital Samson end Scorpion helicopter assets into the background.

John Rosengrant poses with the full-scale Ampsuit After working our details such as weaponry and the configuration of hand controls inside the cockpit, Stan Winston Studio artists rebuilt the original low-rez model in XSI, and then rapid-prototyped the Ampsuit out of foam. Artists detailed the foam pieces and then molded each of the 300 parts separately and cast diem in fiberglass, aluminum and steel. The final Ampsuit stood 13½ feet tall and weighed 1,200 pounds. For medium and close shots of actors at the controls, production often removed the Ampsuit's arms and legs and mounted the torso on a motion platform.

The motion capture cameras themselves were commercial devices for which Giant Studios had written custom integration code. "The cameras are called Smart Cameras," explained Giant Studios motion capture supervisor Matt Madden, "which are like small computers that can be remotely adjusted from our capture workstations via Ethernet connection. We could log into a camera from a main computer and tell it how to interpret an image. So if we had to make an internal adjustment to the image processing, instead of physically having to go to the camera, we could log into it and make the adjustment that way."

Glenn Derry and his crew also set up HD cameras and monitors on the Playa Vista stage. "We were one of the first productions to adopt a high-definition workflow all around," observed Derry. "It has become the standard since we started this picture, but we were beta-testing all this HD equipment that wasn't even on line at that time. We bit the bullet and built the infrastructure around a high-definition workflow; and so all of our video assist, all of our monitors were high definition." Derry devised synchronization tools for the HD, motion capture and facial capture cameras. "They all had to fire at the same time: the mocap cameras had to sync to the reference cameras; the reference cameras had to sync to the facial cameras. And they all had to be locked with time code."

By September 2006, the Playa Vista setup was sufficiently in place to allow Jim Cameron and Richard Baneham to begin capturing flight path animation specifically for the final battle sequence — which would feature aerial skirmishes between aircraft, banshees and leonopteryx — using quarter- and half-inch-scale models that they could hold in their hands. Ramps built on the stage aided ascents, descents and other scaled camera moves. "It was just broad-stroke blocking," Baneham noted, "but it was very helpful. Instead of Jim having to go through an iterative process of telling animators, 'I want the helicopter to fly from here to there.' we could go in and block out the flight paths ourselves."

"Richie and I were like two little boys playing with these models," Cameron elaborated, "but through that process, we captured the entire aerial battle. This was very early on, and we made a lot of mistakes at first. Day One was the dumbest thing you ever saw. We tried to do Ave helicopters at once with different people flying them, and it wound up looking like a game of Twister because all these bodies were in the way. At that point, we realized that we had to break it apart and do key flight paths, with only two or three models at the same time. Then we'd layer more on top of that. It was like sound recording, where you lay down your rhythm track, and then you lay in the vocal and the guitar. We were doing the same thing, multi-tracking with visuals. And it worked very well."

Production designer Robert Stromberg oversaw the design of Pandora's spectacular rainforest environments, a monumental task that would continue over a span of two years. Stromberg and his team tackled the assignment by first designing the wide, bird's-eye views of Pandora, and then closing in, incrementally, on more specific settings and plant life. Final CG forest environments, as created by Weta Digital, were extremely dense and detailed, presenting the effects studio with lighting and rendering challenges beyond any it had previously encountered.

The coarse flight paths motion captured with the handheld models were then fine-tuned through plat-forming. "We could do our initial dive and fly around, for example," said Cameron, "and then we'd fine-tune our banks by plat forming to that move, re-performing the bank angles as we watched ourselves on the virtual monitor. All of these flight paths were based on the kind of coupled motion that you get when an aircraft turns. I've had some training flying helicopters — you don't want to fly with me, but I can do it —and I know how they work. A rotorcraft, especially, doesn't just move along its path through space. It will diverge from its path when it turns, and then correct and come back to it. But keyframe helicopters never do that, which is why they always look like crap."

Cameron and Baneham also captured banshee aerials with models, training themselves to walk the flight path at a constant rate to maintain airspeed. "We could vary the airspeed with the rotorcraft because they can pull up into a hover," Cameron noted, "but the banshees couldn't hover. Banshee flight was based on airplanes, so they had to have constant air-speed. I wound up doing a count in my head to keep my feet moving correctly. There was quite a learning curve forall this stuff."

In the six months leading up to production, the filmmakers also had planned to work with stunt performers to block out big action sequences, with Cameron manning the virtual camera to set up his shots in the virtual realm. "I said to Jon Landau early on." recalled Cameron, "Let's eliminate the previz phase, sitting around in offices, imagining what it's going to be like. Let's eliminate storyboards. We have this technology — let's use it! Let's spend six months before we get the cast just doing capture with doubles, roughing in scenes, figuring out the angles, moving the set pieces around, getting the terrain to fit.' The volume was this free, creative, giant sandbox where we could bring together performers and set pieces and move them around. In a day we could go through three or four scenes with these doubles, blocking out the action, figuring out shots, and getting all the terrain, the mountains, the plants and the skies where we wanted them. Much later, when we came back with the actors, all of that stuff we'd worked out and set up would be there." The pre-shoot blocking period did not work quite as Cameron had envisioned it, however, largely because the crew was still solving a number of virtual production technical problems. "We wound up spending most of that six months just figuring out how to do this," said Cameron, "and so we never really got to pre-capture the movie the way we were supposed to. But if I ever do this again, I'm going to insist that we pre-capture the entire picture. Then, when we bring the actors in, we will be able to go straight to performance. We'll already know where they are going to stand, and where the fight's going to take place, and where the logs are, and where the trees are, and all that sort of thing. And it will be great for the actors, because they don't really care about all of that stuff, anyway. They care about what the character is feeling in that moment."

Jake gets his first good look at the world outside Hell's Gate in an early scene in which he accompanies Grace's science team, in their avatar forms, into the rainforest to collect plant samples. The party is delivered to the site by a Samson helicopter. The Samson, featured throughout the film, was constructed as a 20-foot-long practical prop by Weta Workshop for some interactive shots, out was most often realized digitally by Weta Digital. Weta animators did early motion tests for the Samson and other aircraft — such as the smaller Scorpion helicopters and the Dragon gunship — to develop vertical lift-offs and high-speed flight animation.

As the virtual production crew continued hammering out problems at Playa Vista, Weta Workshop and Stan Winston Studio — now Legacy Effects — joined the Avatar project to elaborate on what had been achieved by the in-house art department. Under the supervision of Richard Taylor, artists at Weta Workshop lent their talents specifically to the development of Na'vi culture and its accouterments. "When we take on a film," explained Taylor, "we like to build the world from the inside out and figure out what the culture is. We want to know the history of a people, and what their religious beliefs are. Are they a matriarchal or a patriarchal society? Do they have a written language or are they working in totem? What do they eat? How would they hunt? And so on."

Guided by the answers to these and many other questions, Weta Workshop would spend two full years designing and building Na'vi weapons, living quarters and clothing, all within the design parameters that had already been set. "For example," said Taylor, "we knew from the original concept illustrations that there were not going to be a lot of feathers on their creatures; and so we had to find something else with which to fletch the Na'vi arrows, drawing on Pandora plant life we'd seen in the artists' paintings. Using modern-day 3D printing and laser-cutting technology, we created highly believable pieces of organic matter that could have, in theory, come from Pandora." Though all the weaponry and other Na'vi props would be created digitally, Weta Workshop most often built physical representations of them — both to get a more visceral understanding of the item, and also to fulfill Jim Cameron's mandate that there should be nothing in the Na'vi virtual world that could not be built in the real one.

Initially, Cameron encouraged concept artists to design bizarre and exotic plants and trees with an overall cyan tint reminiscent of Na'vi skin color. As designs evolved, however, the filmmaker and his artists introduced more green into the environments to create a sense of familiarity for the audience, though many fantastic plant forms remained in the film. Weta Digital built high-rez geometry of approximately 1,500 specimens of plants and trees and, with unique dressing for each, wound up with a library of about 3,000 flora models.

Weta Workshop also built dream catcher-like looms that presumably contained, within their elaborate weaves, the family histories of the Na'vi — a Workshop concept enthusiastically approved by Cameron. "The idea is that Na'vi women weave these long strips that hang in the trees," said Taylor, "and as they feed them through their hands, almost like Braille, they feel the textures, the coarseness and softness of different fibers and little bits of plant life and seeds, all telling the story of a particular family's life and history."

The construction of Na'vi costuming was primarily the work of a young New Zealand fashion designer named Claire Prebble, who had come to work at Weta Workshop after winning an annual exhibition called The World of Wearable Art,' for which Richard Taylor had been a judge. "Claire's fine weaving and beautiful bead work became the core of the Na'vi costuming," Taylor commented.

Weta Workshop also provided practical costuming and props for the Hell's Gate military forces, such as an apparatus that would enable soldiers to breathe Pandora's toxic atmosphere. "We had to figure out what kind of safety equipment these humans would need on this planet," said Taylor, "such as flak jackets and breathing masks, and then design and build those things. We also had to figure out what kinds of things they would need to clear this land and to mine, and then design those pieces of equipment. We did all the military guns, too, from sidearms to full machine guns, going to great lengths to explore the way hardware technology would develop in the future." Workshop artists created a modular system for the guns, which would enable them to transform a simple submachine gun into a grenade launcher simply by clipping together modular components.

Weta Workshop built a number of set pieces for the New Zealand live-action shoot, as well, such as the link unit that controllers enter to connect to their avatars. Working from Rick Carter's design, Weta modeled the link unit in 3D, and from that digital model made a 1/6-scale working prototype that Taylor took to the United States for Cameron's approval. "When we got back to New Zealand," related Taylor, "we used that model to figure out all of the very complex mechanisms needed to make the thing work. It had to open and close at different speeds, while sliding in and out of a rotating ring, hiding all the components within a very tight canopy. It was a tricky mechanical engineering job, requiring large pneumatic pistons to drive the system." Weta Workshop built five of the link units, plus two of the huge tubular chambers from which the avatars are born. "This wasn't just model making. It took a year to build those two units because they were very complex and difficult. When Jim first came and saw them, he walked around them, asked us a few questions, challenged us on a few decisions — and then absolutely loved them. When you've put that much time and labor into something, it is very gratifying to please the director." Weta Workshop built a full-scale fiberglass exterior of the Samson helicopter, as well, which measured 20 feet long by 12 feet high. Workshop artists Leri Greer and Aaron Beck also designed graphics seen on the screens and monitors in the Hell's Gate control center.

Grace's avatar was the most difficult to realize, largely due to the incompatibility of Sigourney Weaver's face with the features of the Na'vi. Weaver's narrow, patrician nose, in particular, did not translate well to the broad, leonine Na'vi nose, characteristic of the avatars. In addition Grace's avatar had to appear as an 18-years-younger version of the character and the actress. Stan Winston Studio tackled both issues, referencing photographs of Weaver as she appeared in Alien 3 to design a younger avatar face, with a nose that more closely matched her own.

Half a world away, Stan Winston Studio effects supervisor John Rosengrant was overseeing artists Chris Swift, Joe Pepe and others as they detailed Na'vi and avatar designs in Photoshop, developing final color and paint schemes for Cameron's approval. The SWS team created variation after variation of these paint schemes, offering Cameron as many as 25 iterations of blue skin, for example, with extremely subtle differences between each. "One would be just a little more blue," noted Rosengrant, "another more gray. The battle was coming up with a blue that would still look believable as flesh. There are blue-colored fish and birds, but no mammals with blue flesh that we could reference." SWS also presented Cameron with Photoshop renderings of different stripe and bioluminescence patterns that would appear on the characters' bodies. "For the stripe patterns, we referenced tigers and zebra. We actually superimposed zebra stripes on our Na'vi at one point. After we came up with many of these stripe and bioluminescence patterns in Photoshop, we presented all those options to Jim, and he narrowed down what he wanted for the Na'vi."

Once Photoshop renderings had been approved, SWS took photographs of people striking poses and making various expressions: and then artist Scott Patton brought those photographs into ZBrush to resculpt them as Na'vi characters. "We got into the habit of retaining the mouth area of the person in the photograph, and blending that into the face," said Rosengrant. "Ultimately, it would be the motion captured actor's mouth driving the animation; and so, to have a Na'vi mouth area that was true to the actor would be helpful. We also retained the area around the eyes of the people in the photographs — again, to help translate the actors' expressions to their characters. The Na'vi eyes are very large and yellow, very catlike; but we kept the area around those eyes much like it was in the photographs. The brow shape was also cat-like, but we mixed it with the real person's brow." Once specific actors had been cast in Na'vi roles, SWS repeated the technique, importing photographs of the actors into ZBrush and sculpting Na'vi features on top of those images. "Finding the right look for Neytiri was especially critical because if she didn't work, if you weren't attracted to her, the whole story wouldn't work."

Though Neytiri is pure Na'vi, without the DNA and features of a human controller, the character wound up resembling Zoe Saldana to a degree —just as the matriarch, Mo'at, would resemble CCH Pounder and the Na'vi patriarch, Eytukan, would resemble actor Wes Studi. "We found that the best way to convey the essence of a performance, to see if we were really getting an actor's performance, was to bring in similarities in the faces of their Na'vi counterparts," observed Jon Landau. "We never see the Na'vi characters as human in the film, of course, so you never have that reference poi nt. But if you know CCH Pounder, you recognize her as the matriarch. You recognize Zoe Saldana. You can definitely see them in those characters."

Avatar sculpts incorporated even more features of their human counterparts to ensure they would be recognizable as the actors. Of all the avatars, Grace was the most difficult to realize, as it had to represent a hybrid of a Na'vi and a much younger Sigourney Weaver to accommodate the story point that Grace's avatar had been created 18 years in the past. To work off reference of Weaver at the appropriate age, SWS brought pictures of the actress as she had appeared in Alien 3 into ZBrush, and sculpted the avatar features from those photos. Weaver's features — at any age — also turned out to be the least adaptable to the Na'vi face. "The Na'vi facial features include this very broad, lion-like nose," commented Jim Cameron. "It's like a brick in the middle of the face. They also have a forward-projecting brow and very high cheekbones. All of that worked well with Sam Worthington's face. And some of it worked fine with Sigourney's face — she's got the high cheekbones and the strong jaw. But she also has a very narrow, patrician nose; and the second we tried to put that wide Na'vi nose on her, you lost her. You couldn't tell it was Sigourney Weaver anymore. So we had to go through many iterations with John Rosengrant and his team, and they were fantastic. With their years of experience in makeup design, they knew exactly what to do, which was to make Grace's avatar without the wide nose. We just gave Grace's avatar Sigourney's nose, and it worked fine because the character still had the big eyes and other features."

Jake encounters a hammerhead titanomere in the forest. Concept artists designed the rhinoceros-like animal with anvil-shaped horns, and then painted intricate markings to cover the creature's body. Early on, Cameron had conceived all of the terrain-based creatures as having six, rather than four, legs, before committing to the notion, however, he hired ILM animator Paul Kavanagh to create a series of walk and run cycles to determine if a six-legged creature could move gracefully and powerfully, or if the extra legs would inevitably create on awkward gait. The animation cycles proved the viability of the idea, and all of Pandoras land animals were endowed with six limbs.

To give Cameron something more tangible than a 3D image — something he could touch and look at in different lighting scenarios — SWS created human-sized busts of the main Na'vi and avatar characters, based on scans or traditional lifecasts of the principal actors. "Looking at a computer character in the 3D world just isn't the same as being able to hold a model and put a light on it," said Rosengrant, "and so we took a lifecast of the actor, did a clay press-out and sculpted that into a Na'vi or avatar character. These sculptures were full-scale in that they were human-sized, but they were not true Na'vi-sized. By making them human-sized, we could easily retain the actor's features from the lifecast. We also created some full-body, human-scaled models."

Later, SWS rapid-prototyped full-scale Na'vi models that stood ten and a half feet tall. "These were generic male and female Na'vi models," commented Rosengrant, "something they could have on set to remind them how tall the characters would be in shots." The full-scale models were dressed and adorned with the costumes and jewelry that had been designed and fashioned by Weta Workshop. Finished models were scanned to generate 3D data that could be turned over to Weta Digital for the building of generic high-rez Na'vi characters.

In the same sequence, Jake outruns a ferocious thanator, primarily a Jim Cameron design. Animation supervisor Richard Baneham developed rigs to explore the animal's cat-like behavior, which was more fully developed by the animators at Weta Digital. To guide chose choreography, production provided Weta with motion capture of a stuntman, playing Jake, running over and under obstacles placed on stage — but thanator action, such as lunges, leaps and swipes, was left to animators'imaginations. / Jake defends himself against Pandora's deadly wildlife.

The Winston team also contributed design elements for creatures such as the viperwolf, rendered in Photoshop by Jim Charmatz. Scott Patton then sculpted the intricate head of the beast in ZBrush. "The viperwolf is where we really got into 3D sculpting," Rosengrant stated. "We could build a 3D model in ZBrush and put it through its paces — and we could do it with Jim. on the fly. We had two-and three-hour sessions with Jim during this process, working on characters with him on the computer." The full-body design of the viperwolf was sculpted by Jason Matthews, done old-school-style in clay.

In addition to designing characters and creatures, SWS built a practical version of the Ampsuit for live-action shots of the Hell's Gate armor bay and the final battle, working from the detailed 3D design model TyRuben Ellingson had created with Jim Cameron. To produce a full-size mockup — something the Winston team had done frequently ever since it had been assigned the full-size alien queen for Cameron's Aliens — the artists assembled rough foam core and wood pieces into an Ampsuit form. Rosengrant and SWS model shop supervisor Dave Merritt then took the mockup to Playa Vista to demonstrate it for Cameron, operating movements using telemetry suits from past projects.

After working out more details, such as the look and function of hand controls inside the cockpit and the suit's weaponry. SWS remodeled the original digital model in XSI, and then rapid-prototyped the Ampsuit out of foam. "We got these simple foam pieces back," stated Rosengrant. "and detailed them, carving in seam lines, sinking in hardware and unique-shaped bolts, and putting texture on them based on reference of Abrams tanks and Apache helicopters. And then we started molding each part. There were around 300 parts, maybe more. We cast up all of the parts, mostly out of fiberglass, with as much aluminum structure as we could get away with in order to keep the weight down. There was also steel in it to help maintain the integrity." SWS fabricated five different Ampsuit canopies in various states of damage that could be interchanged, depending on the scene being shot.

The final Ampsuit stood 13½ feet tall and weighed 1,200 pounds, about 800 pounds of which was the torso. "It was completely posable," said Rosengrant, "and we could have loosened the arms to rod-puppet them, as well — although we never did that. During the shoot back i n Los Angeles, we took off its legs and mounted it to a motion base and moved it around while the actor was in there, and CG arms were added later. Glenn Derry and his team provided the motion base, programmed to reproduce the movements that Jim had set up in his motion capture shots." Dave Merritt headed up the Ampsuit build; Ken Coronet was the lead model maker and maintained the suit during the shooting of live-action Ampsuit scenes in New Zealand and Los Angeles.

The virtual production crew ran tests with the actors to calibrate their bodies to the performance capture system in early 2007; and then, in April, Cameron began shooting Avatar on the stage at Playa Vista. Typically, the production process began with the virtual art department pulling environmental assets, based on concept paintings, for scenes on the upcoming shot schedule. First assistant director Josh McLaglen would then do a pre-scout, studying the raw set from the virtual art department to make certain it fit the requirements of the scene. "The first few times I came in." recalled Cameron, "we spent half the day just getting in all the right set pieces so I could actually work. So finally I said: 'Okay, let's do a pre-scout first. Just get us off stupid and make sure that the set the virtual art department has created fits the minimum requirements of the scene. If the scene calls for a waterfall, make sure there's a waterfall. If it calls for some specific action, make sure whatever we need for that is there.' That was step one."

When the virtual art department had gathered all the requisite set pieces, props and character assets, the motion capture crew would run a test to ensure the virtual scene was not too dense for the system's real-time playback capabilities. "The more we put into a scene," Nolan Murtha explained, "the lower the possibility of real-time became; and so we needed a test to make sure we'd be able to capture that scene in real time."

Virtual environments were then sent to the stage crew, which would assemble physical representations of hills, logs, trees, vines and other environmental elements in the scene. "We created stage sets that would match all the virtual sets." Cameron said. "We were working in this flat motion capture volume, but nothing in these jungles or mountains was flat. There were ravines, rockout croppings, slopes, and all kinds of things. So we had to create physical sets that could support the actors and map very quickly to our virtual terrain. Nobody really knew how to do it at first, and it stopped us cold for a while. Finally, I said: 'Let's just divide the volume into six-foot grids. Each one of those six-foot squares will have a riser on casters, with terrain pieces built onto them. And each of those risers will lock together with other risers like a big jigsaw puzzle.' They took that idea further and used an alphanumeric coding system for these six-foot grid forms. So, for example, the 'Scene 41A' grid would be a particular configuration of these six-foot risers with all their little polygonal dressing shapes on top of them. We found that we could shoot on one grid, disassemble it, bring the next grid in and be ready to shoot the next scene in fifteen minutes. Once it had been fully designed, we could hopscotch through the whole movie with this grid system, moving from one scene to another." Built onto the risers were not only representations of terrain matching the digital environment, but also specific set pieces with which characters would come into contact. "We'd start with the terrain grid, and then bring in non-occluding set pieces for contact alignments — the vine that someone would be using to pull themselves up, or a log that someone was going to jump over."

Working on the assembled stage set, Cameron would first block out a scene with the virtual camera and a troupe of stunt performers. Conducted without benefit of previz or storyboards — these sessions were in essence the previz and story-boards — the blocking sessions were a freeform exploration of camera and action. "We were basically making it up as we went along," said Cameron, "which worked out fine. There was no detriment to the end result; it was just a little chaotic, mainly because we hadn't been able to get the technology on line in time to do the pre-capture as I'd wanted. And then we were up against deadlines, and we had to start working with the cast. But we never lost this idea of roughing it in with doubles. We never went into a scene with the actors without first blocking things out and setting up shots with the stunt performers. That little troupe of six or seven people proved to be absolutely critical. They were all just remarkable physical performers, all stunt people who could also act, who wound up playing all the different roles for these purposes. Kevin Dorman played every character in the film atone point or another — Quaritch, Jake, Mo'at, Neytiri. We wanted to put him up for the SAG ensemble award!"

Most often, Cameron would film the actors, garbed in their mocap suits and head rigs, within a few days of the initial scout and blocking — although, that shoot could also take place much later. "Usually, we'd scout maybe ten days' worth of capture." Cameron noted, "and then we'd do that ten days of capture with the actors. Then we'd go back to the editing room, cut it all, get it turned over to Weta, and go out and scout for another few days."

Neytiri strikes a warrior pose. Some shots of CG characters were achieved by blending the emotive facial performance of the actor with the dynamic body performance of one of seven stunt people that stood in for actors and even portrayed creatures for mocap purposes throughout production. Neytiri's lithe, yet powerful athleticism was central to her characterization, as was her sexual appeal — despite cat-like yellow eyes, pointed teeth and blue skin.

With the actors on stage, Cameron would often start by framing up shots, using the virtual camera as a viewfinder, a window into the virtual world. "Jim would find his shots with the virtual camera in hand," explained Nolan Murtha, "looking at the virtual reality environment. Even though he would usually go back and add camera later, after the actors were gone, it was imperative that he be able to see that virtual reality live as he set up shots."

After setting up his shots, Cameron would concentrate on the performances and HD camera coverage. Eight to twelve HD cameras were in the volume with Cameron, filming the actors from a variety of angles, with the virtual camera acting as the 'A' camera to set the scene's axis. "We'd shoot a closeup, a medium and a wide from complementary angles to the main camera," explained Richard Baneham. "So if we had a scene with two actors in it, we would have a wide shot that showed their interaction, a medium-wide that showed both of them relative to each other, usually from the knees up, and then dedicated closeups. If there was intricate hand detail or an exchange of a prop or something like that, we would also have a dedicated camera on that."

Though the main purpose of shooting the HD camera coverage was to provide invaluable performance reference to Weta Digital's animators, it also served to create a filming environment that was very much like that in alive-action film. The director was shooting his actors, framing them and getting a sense of what the final shots would be; and, as a result, the actors knew what their relationship to the camera was, and what type of shot it would be when Cameron finally added camera to the digital version of their scene. "Actors act differently depending on the camera," noted Murtha. "If you're doing a wide master of a scene, they're going to be a little more grand in their motion than they would be in a closeup. The problem with shooting traditional motion capture and then adding all the camera later, which is how it has always been done, is that the talent doesn't know when they are in closeup or when they are in a wide shot. It's all the same for them on the motion capture stage. So you lose all those subtle differences of how an actor might act in a wide shot as opposed to a closeup. And it isn't valid to just tell them. 'Okay, act as if this is a closeup, even though the camera is 50 feet away.' That doesn't really resonate with them, and so you're not going to get the same type of performance."

During the shoot, the head-rig facial camera footage was mapped to the faces of the digital characters displayed on the virtual camera and stage monitors — a kind of kabuki mask technique that stood in for facial animation that would not be forthcoming from Weta Digital for many months. "Instead of waiting to get an actual articulated face," explained Baneham, "we just built a piece of geometry that was about the same shape as the face, but without all the detailed forms; and then we projected the head rig camera video stream of the actor onto this face shape. This was a way to get an idea of the facial performance and see the emotional content right away. It was an aid in helping Jim as a director and a camera operator because, even with just this simple kabuki mask, the emotions came through."

Distrustful of the avatar intruder, Neytiri is prepared to kill Jake with her bow and arrow, but hesitates when glowing woodsprites surround him — an event she interprets as Nature's gentle nod of approval. Pandora's many glowing, bioluminescent creatures and plants were inspired by underwater organisms Cameron had observed during countless diving expeditions. To suggest the bioluminescent effect, as seen at night, concept artists drew scenes in neon-colored pencil on black paper, a technique Cameron had used to illustrate the look he wanted for the radiant aliens and their craft in The Abyss.

After a performance capture session, Cameron would do a rough cut of the HD footage. That rough cut would then be disseminated to all the virtual production departments so that they could prepare for a virtual camera session — the point at which Cameron would actually add camera. Prepping of a scene included a refinement of the motion capture by Giant Studios. "We'd provide Giant with the selects and time code information," related Murtha, "and their trackers would remove any poppiness or erratic behavior. Then we would rebuild the scene with the refined motion. Giant's motion capture system is the best in the world; but if for some reason the computers couldn't regenerate the motion, we would hand over all the HD reference to Giant, and they could rebuild things based on that reference material, putting legs where they needed to go and that type of thing. They weren't animating the shots, they were just reconstructing what should have been there."

In the virtual production lab, crew members used the HD rough cut to determine which CG characters and environments would have to be pulled for the virtual camera session. "If the 3D art department could see that Jim would be covering a shot from one side," noted Matt Madden, "they didn't have to worry about the details of the environment on the other side. The environment was so rich and dense and detailed, anything they could do to narrow down what was needed for camera was helpful."

Cameron would then review the assembled scene in digital form, and add elements or change things as needed. "Jim would say, 'I need more fire over there,' or 'I need the helicopter retimed,'" said Murtha. "And we were able to do that with only a couple of days' turnaround for several shots or a whole sequence — because what we were looking at during the review process was the MotionBuilder template: the CG characters with the cleaned-up motion capture applied to them, in a CG environment, all in low-rez form. We could swap out a whole environment or change a scene from night to day very quickly because we were working with such low-resolution assets. Our render time in MotionBuilder was a half hour for a very long shot, whereas that same shot coming out of Weta would take hundreds of hours to render."

After environments and backgrounds had been dressed and tweaked to Cameron's satisfaction, he would add camera, using the virtual camera to move through the CG scene. "Jim would shoot coverage on his performance selects," said Matt Madden, "and the live output from the virtual camera would be streamed into the Avid, where there was an editor, assembling as they went. Occasionally, Jim would walk over and sit with the editor and see how things were cutting together as a new scene was being assembled and loaded. He could tweak timings during that few minutes of downtime, waiting for the next load. Once the new scene was loaded and he felt good about the previous shots, he'd go on to the next camera move."

The final scene, edited to its precise length and with camera added, was called the template, and was subsequently turned over to Weta Digital. "We gave Weta what was essentially a rough videogame version of the scene," observed Nolan Murtha, "complete with effects and environments, mist and clouds, characters. Everything that was going to be in the final image was there in this videogame-resolution version that Jim had shot his camera work to."

One of the creative challenges for Weta Digital was determining how the bioluminescence would be introduced as Joke embarked on his night in the forest. Early tests had the glow appearing suddenly — at if turned on by a light switch — as Jake's torch was extinguished. That concept gave way to a more gradual introduction, the bioluminescence subtly dialed up throughout the night scene to avoid distracting from the action — until Jake himself stops and takes wondrous note of his glowing surroundings. Weta created the final look on a shot-by-shot basis, using a variety of 2D and 3D techniques.

Along with the template, Weta was also provided with additional facial performance reference. To augment data and reference captured by the head rig camera and the HD cameras, the filmmakers had conducted dedicated facial capture sessions that they dubbed Facial Performance Replacement, or FPR — a conceptual spin-off of Automated Dialogue Replacement, or ADR. "If an actor can watch his performance and change his vocal performance later," Cameron inquired, "then why not do the same thing with facial capture? Particularly in very active scenes, we found that it was difficult to get the reference cameras to follow the actors while they were jumping around. So, instead, we'd just settle them down in front of a monitor and have them watch either themselves or the stunt double, and take the timing for their expressions and lines off that physical performance. They would do their facial performance and say their lines as cameras captured them from different angles, in closeup. So we were uncoupling the facial performance from the physical performance, but in a way that the actors embraced. It actually freed them up to perfect their performance without having to worry about how they were jumping or rolling around. It worked beautifully, and it gave us great reference angles where we could see the actor's face very clearly."

To guide their creature animation, Weta animators also received motion capture data that had been acquired as members of the stunt performer troupe performed basic creature action. That motion capture data was subsequently mapped to a CG creature rig. Like the flight path blocking Jim Cameron and Richard Baneham had performed in the early days of the production, the creature mocap was merely a blocking tool, a way to improve communication with Weta's animators. "It was very crudely laid in," said Baneham, "but it allowed us to say: 'This is what we need. The thanator starts on the left, crosses to the right, and there is an impact in between.' With all of that basic blocking laid out, Andy Jones and Dave Clayton and their animators could get their teeth into the scene that much faster and could spend their time where it was warranted, creating really high-end animation."

Another advantage to motion capturing the performers as stand-ins for CG creatures was that it created a sense of timing and compositional space in the template files sent to Weta. "As an animator," Baneham commented, "I've found, consistently, that I've had to fight for screen space and screen time for a creature that's not there when they shoot the plates. It makes sense — camera operators shoot what they can see, and editors cut to what is there in the plates. But often we would end up trying to put an animated creature into too short a time or too small a space. By doing this motion capture with performers, we were able to compose the shots around the creature, as if the thanator were really there on set. The creature performances were grounded in reality. So much about this movie was sublimely surreal, any time we could ground the audience in reality, it would help to keep them in the movie."

Jake briefs Quaritch on what he has learned about the Na'vi in the Hell's Gate operations center, where the military keeps watch on the native people and their Home Tree via holographic surveillance. Frantic Films produced the 'holo-table' shots, first executing a rigorous tracking of the live-action photography and then creating tree models from production reference imagery. Frantic used Krakatoo software to generate particles on the surface of the tree geometry to create a lidar-style look. Frantic also designed the graphics and icons seen on the ops center monitors.

The creature-actor mocap also lent reality to scenes in which creatures physically interacted with Na'vi and avatar characters. For the scene in which the viper-wolves chase Jake through the forest, Cameron, Baneham and a couple of stunt performers stood in for the viperwolf pack, carrying gym bags at their sides to approximate the height of the predator in relation to Sam Worthington. "These gym bags were stuffed full of material so they had weight to them," said Baneham, "and they also had calibrators strapped to them, patterns that the mocap system could recognize. We ran around with these gym bags, as wolves, chasing Sam through the set. It was very physical, very interactive. At one point, a viperwolf knocks Jake down — and we really knocked down Sam; so we had that impact, that weight, right there in the performance, which really helped to lift the level of animation. When you see it in the movie, you can tell that it is really a guy being knocked down."

Production sent the creature mocap files to Weta Digital, where the animation team fully animated high-rez models and then sent them back to the stage in the form of low-polygonal assets so Cameron and his crew could shoot cameras on the scene. "Jim couldn't shoot final cameras until he saw what the real creature animation was going to be," Baneham noted.

Using performers to choreograph creature blocking was an aid in realizing one of the film's most dynamic scenes — the final-battle fight between Neytiri atop a thanator and Quaritch in an Ampsuit. "We could have keyframed this fight," stated Jim Cameron, "but instead I choreographed the whole thing with two performers so it would have a very natural feel to it. We weren't even worried about the camera at that point, just the choreography." Cameron motion-captured stunt performer Alicia Vela-Bailey as the thanator, wearing a small Neytiri sock puppet on her back to help her remain cognizant of the character as she performed jumping and rolling moves. In a second session. Vela-Bailey portrayed Neytiri atop the thanator. "We just put her on this thanator back, and, by hand, we moved it violently and tried to throw her off, like the bull at Gilley's. That gave us information about what Neytiri's hair would do and how she'd move as she tried to hold on."

A holographic display of a consortium mining pit. Frantic Films modeled the interior of the holotable and positioned it within tracked scenes. For the animation and design of screen graphics — air traffic control-type displays, as well as views from security and vehicle-mounted cameras — Frantic Films adhered to the production brief, which was to create a sense of depth in the imagery. Frantic Films created graphics in Photoshop and After Effects, and then ran them through a proprietary tool to generate stereo layers for compositing in Fusion. Frantic also added buildings, aircraft and ground vehicles to create outside views seen through the ops center windows.

Cameron blocked out the Quaritch and Ampsuit action in a similar fashion, with Stephen Lang playing Quaritch and stunt performer Kevin Dorman playing the Ampsuit in a tandem performance capture session. That captured motion was then used to drive the practical Ampsuit on a motion base when Cameron shot photography of Lang inside the suit, working the controls.

The crew also captured actors or stunt performers on gimbal rigs for aerial scenes in which characters were piloting aircraft or riding atop flying creatures. To guide the gimbal operators, the crew displayed the flight path animation captured by Cameron and Baneham for each specific shot. "We'd put a tail camera on the flight capture," said Cameron, "so we could see it on this monitor. We could see where we had a right bank, a sustain, a left bank and dive, a right bank and climb — you could see it all on the monitor. It was like riding a rollercoaster, and we could anticipate what was coming. So Richie and I would stand there with our arms outstretched, watching this monitor, and we'd anticipate what the next move was going to be by the amount of time it took the gimbal crew to react. Right before we started to go into a bank, I'd demonstrate it with my arms, and they would hit the bank at the right time. And we could actually see the actor performing the character in real time, virtually, sitting on the creature, for example, as it went into a turn. It was kind of a mess at first, but we got better and better at synchronizing the motion."

To heighten the actor's sense of executing banking and other flight moves, stunt coordinator Garrett Warren would push down on the gimbal wing and push up on the actor at the moment the bank began. "The actors would feel that movement and react to it naturally," explained Cameron. "They would thrust down with their downhill leg, and throw their center of gravity uphill. It created this nice, athletic, jockey look — not sitting low in the saddle, but perched up on the pegs, floating dynamically."

In a beautifully lyrical sequence, Jake and Neytiri abandon themselves to a joyful flight perched atop their banshees, circling, diving, climbing — all without taking their eyes off each other. "This is the moment where you feel Jake and Neytiri falling in love," noted Cameron, "and they're not even standing together. They're on separate creatures. But you see this dance between them that has an emotional component to it. So, of course, we had to deal with eye lines for that. With all of these flight moves, how would they know where to look?" The actors rehearsed the sequence looking into each other's eyes first, to find their way through it emotionally. "Then we put a day-glo tennis ball at the end of a long extendable rod, and the actor would take that same emotional idea, but play it to the tennis ball." The actor on the gimbal would keep his or her eyes on the tennis ball as the crew maneuvered the rod in accordance to the other character's movements in the flight animation. "We'd be swinging this rod and tennis ball around, saying, 'Okay, here comes Jake; he's going over your head; and now he's over here!"

On the capture stage, actors perform a scene in which Jake and Na'vi tribe members climb a mountain in search of flying banshees. For each performance capture session, Cameron first would explore low-rez environments built by the virtual art department to determine the setting for a scene. Virtual environments were then sent to a stage crew, which assembled practical forms representing hills, logs, trees and other terrain elements, built on six-foot-square platforms with casters for ease of movement on and off the stage. / Actors on platform set pieces appeared as CG characters climbing virtual terrain on monitors set up around the stage.

In late October 2007, after six months of shooting performance capture at Playa Vista, the production relocated to new Zealand to shoot live-action sequences on stage sets. Cameron had made the decision to shoot at Weta during one of his previous visits. "I think Jim was impressed with everything he saw here," stated Joe Letteri, "in terms of stages and the overall facility. He also thought that if he shot here, it would give him the opportunity to be close to us while we were doing a lot of the R&D work for the visual effects. We would be able to just shoot over to the stage and show him things on a laptop." Originally scheduled to last eight weeks and to be wrapped by Christmas, the live-action shoot was extended several weeks into 2008, with Weta Digital visual effects supervisor Stephen Rosenbaum covering the shoot as Joe Letteri continued to oversee the R&D effort.

Cameron photographed all of the live-action with the stereo 3D camera he and Vince Pace had developed, which Pace had refined and reconfigured in the previous months to make it Avatar-ready. "For Jim to do Avatar," said Pace, "all the 3D technology had to be transparent to production. It couldn't in any way impede the process of making the film. So the key factor was to make it the most transparent tool possible from a production standpoint. Of course, making the technology transparent is much easier with a director who works mainly off a crane or a dolly. But Jim has a tendency to push the cinematic style. He likes camera movement. He likes to have the camera on his shoulder right next to an actor as he's talking to him — but then he wants to be able to turn around and fly the camera in on a Technocrane. So we had to make something, that could adapt quickly to all those situations. There couldn't be long gaps of time while we went from a handheld camera to a crane-mounted one." Just two weeks prior to the start of filming. Pace received a lengthy email from Cameron in which he detailed additional items on his camera wish list. "Jim wanted the camera to be able to turn upside-down. He wanted it to have lensing capabilities that didn't exist yet — and here we were, two weeks from the shoot! At that point, we went into serious batten-down-the-hatches mode, and we built the cameras with all those requirements within that two-week window. We had the foundation already, of course, but a lot of shit happened as we got closer to production."

Jake and the Na'vi arrive at a banshee rookery, each intending to claim one of the reptilian creatures through a mutual spiritual connection and then partner with if in an aerial hunt. Animators at Weta Digital produced a series of motion studies to develop nesting and roosting banshee behaviors in the rookery, as well as lift-off and flight animation. Banshee wing movement was determined in part by its bat-like configuration — a leathery membrane stretched between ribbed structures, as mandated by Cameron's edict that all the creatures in Avatar would be without fur or feathers.

The final 3D camera rig, like its predecessors, was actually two cameras — one for the left eye, one for the right — synchronized and mounted at a right angle to each other. Weighing between 25 and 33 pounds, the rig is as mobile and user-friendly as any motion picture camera. "It is really a miracle piece of technology," observed Cameron, "a 28-pound camera that I can put on my shoulder, with a nine-axis motion base that's completely silent — so silent you can do sync dialogue two feet from it. In the nine years leading up to Avatar we had made all these refinements to the camera; and we'd also worked up all the stereoscopic methodologies. So we knew how to manage the stereo space, and we just applied those rules. During the first week of shooting there was some recalibration and rethinking about how to do things; but by the end of that week, we had our methodology dialed in."

Though stereoscopic hardware and techniques were fairly routine for Cameron and his crew by the time production arrived in New Zealand, the live-action shoot was not without its innovations. The most significant — not only to Avatar but also to future film productions — was 'Simulcam,' which executed real-time comps during filming, enabling Cameron to see in his camera viewfinder digital set extensions rather than greenscreen, and digital characters interacting with live environments and actors.

The development of Simulcam had been initiated two years before, as Cameron considered the virtual camera one day and asked Glenn Derry what turned out to be a very loaded question. "I'm standing there with the virtual camera," Cameron recalled, "and I realize it's not a camera at all, it's just an object. The system recognizes it as an object, orients it spatially within the environment, and then assigns it the role of being a camera, and renders an image as if it were a camera. But it's not a camera. It could be a toaster, right? The cognitive leap I made, suddenly, was — what if it really was a camera? What if you superimposed a grid system for the virtual camera onto a motion picture camera that was actually generating an image at the same time? Couldn't you merge the two worlds?

In addition to the aerial hunt sequence, the banshee would be featured in a romantic flight scene with Jake and Neytiri and in the epic final battle. Concept artists Wayne Barlowe, Neville Page and Daphne Yap rendered banshee concepts with the aim of finding a new take on a flying reptile, a staple of fantasy films. Lead designer Yuri Bartoli continued to tweak the banshee design for many months, pointing high-resolution treatments in Photoshop to develop detailed textures, colors and markings. Bartoli used ZBrush to project the textures onto a banshee model and determine how the 2D concepts would translate in 3D.

"So I asked Glenn Derry: 'If our virtual camera was a camera and was photographing actors, and there was a greenscreen behind them, couldn't we feed in the image generated by MotionBuilder, and have a real-time track that allowed us to either extend the sets with virtual environments or put virtual characters into it? And couldn't those virtual characters interact in real time with live-action photography?' Glenn thought about it for a second, and said, 'I can't think of a reason why that wouldn't work.' Of course, there were a hundred reasons, and it took him two years to build it. But from that moment, we were on. We had our capture methodology. We had our live-action methodology. We put the two together and called it 'Simulcam.'"

Simulcam required that Derry and his crew build a mobile motion capture system that could travel relatively easily from one live-action set to another and adapt to a variety of setups. The team also had to grapple with the issue of doing motion capture within various live-action lighting environments. "Motion capture volumes are clinical, sterile places where you have very controlled lighting and everything is built around the parameters of the motion capture system," said Derry. "But on a set, instead of those sterile lighting conditions that are optimal to the function of the mocap system's infrared cameras, we'd have 20K or 50K xenon lights interfering with the motion capture."

"The lights spoof the system because it sees any light source as a marker," Cameron elaborated. "So Glenn had to come up with a light-proof capture system. It would use the same Giant Studios' system on the back end, taking the data into MotionBuilder exactly the way we had been doing it; but Glenn had to redo the actual capture marker system completely from scratch."

What Derry came up with was an active light emitter that was synchronized to the shutter speed on the mocap cameras in such a way that it eliminated interference from the set lighting. "On both the cameras and the suits worn by the actors," Derry explained, "we put these little LEDs that would fire off extremely brightly in phase with the motion capture camera. We would fire off these LEDs for 20 microseconds; and so, all the motion capture camera would see was this dot. You could point it at the sun, and you wouldn't see the sun — you'd just see this dot. I knew if I could beat the sun, I could beat the 50K light on set."

Stuntman Terry Notary performs banshee behavior on the motion capture stage. Production captured stunt people as creatures, as a means of determining spatial relationships and working out basic action and blocking. After stunt performers had been captured, the data was mapped to a CG creature rig by the virtual production team and sent to Weta Digital. Though the captured animation was crude and lacked the dynamics of a thanator's cat-like leap or a banshee's graceful flight moves, it gave Weta animators solid information regarding creature choreography in a scene and allowed them to concentrate their efforts on creating high-end animation.

The final motion capture unit was mounted on casters so it could be wheeled from set to set, and also had mounts that allowed it to be attached to sets in any number of ways, usually from tall towers. Since the motion capture cameras would be constantly changing positions — unlike the ones mounted permanently to the ceiling at Playa Vista — Giant Studios had to write new calibration software. "There was no way for us to know exactly where the motion capture cameras would be positioned in a set until all the lighting setups were done and all the set pieces were in place," said Matt Madden. "So we needed the flexibility to change the motion capture camera setup on the fly and be ready to roll live within minutes. To do that, we had to write new tools that allowed us to recalibrate the cameras very quickly."

Typically, the crew would capture the Na'vi or avatar actor performance for a scene first, and then map that performance to the digital character when Cameron shot the live-action. "Capturing that performance first gave Jim a known element," said Derry. "He knew that Sam was going to do exactly the same thing every time; so he could just work with the camera and the other talent in the plates. For the scene where Jake first wakes up in his avatar body and stumbles around the room, Jim spent the morning directing Sam Worthington on a scaled, motion capture set that we built. And then, once we got to the real set, we played back Sam's performance and Jim could line up to Jake in every shot. He saw Jake's avatar through his handheld camera eyepiece as he shot that live-action scene on the set."

To allow Cameron to also see digital extensions to Rick Carter's live-action sets through his camera view-finder, the virtual art department back in Playa Vista sent low-rez set extensions and environments to New Zealand. For example, Carter had built an Operations Center set with greenscreen material in large window areas. The virtual art department provided the main unit with exterior Hell's Gate environments so that Cameron could look into his viewfinder and see the view — at least in low-rez form — outside the windows, which included vehicles and helicopters on a tarmac, perimeter fencing, and the jungle beyond.

Simulcam was an unmitigated success, and Cameron wound up relying on it heavily throughout the live-action shoot. "We became addicted to Simulcam," Cameron said, "We used it for everything, and seven times out of ten it was justified — the rest of the time, we just used it because we wanted to. We used it whenever something in the CG world affected something in the live-action. We used it for the views out the windows. We used it to understand the lighting. We used it to understand composition, especially when the live-action was only a small part of the frame, and the rest of the frame was empty. Normally, it can be very scary to shoot all this greenscreen, with your live-action portion pushed down into the corner. You get into that situation and every instinct you have says 'pan over.' But with Simulcam, you don't, because you can see what the shot's going to be. It allowed us to operate in a very fluid, cinematic way.

"People have been saying that real-time performance capture is going to change filmmaking, that it's revolutionary — but, really, it's only going to affect a small group of people. There aren't going to be that many films that need it, or that many filmmakers who are interested in it. But Simulcam could really revolutionize filmmaking. These days, there are very few films made without visual effects; and most of them are live-action based. Integrating the CG production pipeline with live-action photography, even in 2D, is something that could have broad applications to the industry."

Jake takes his maiden flight on a banshee. Cameron and Richard Baneham blocked out basic banshee flight paths early in production, using models as motion capture subjects. Since banshee flight was based on airplane flight characteristics — other than the hovering motion of a helicopter-type craft — Cameron and Baneham had to simulate a constant air speed as they walked the flight paths on the motion capture stage, using a counting system to maintain their rhythm. Production also motion captured banshee wing action, targeting crew members in mocap gear flapping their arms at appropriate speeds.

With the completion of the New Zealand live-action shoot in early 2008, the filmmakers returned to the stages at Playa Vista, where they would continue to shoot motion capture material and, later, pickups and inserts. Official shot production began at Weta Digital at about the same time — although the company had spent the entire previous year on research and development, with Scene 53, in which Neytiri stalks Jake in the jungle, serving as their testing ground. "It was only twelve shots," said Joe Letteri, "but they were very challenging because they introduced these characters, and we were really seeing the jungle for the first time, as well. So, for these shots, there were a lot of creative decisions that had to be made. How blue should the plant life look? How alien? If it is too fantastic, will the audience be overwhelmed and lose focus on the characters? We had to find the balance between an alien world and a familiar world, and those creative issues took a fair amount of time to figure out."

The other challenge was the purely technical one of lighting and rendering shots that were bigger, in terms of polygonal counts, and more complex than any Weta had ever done, even for the ambitious King Kong or Lord of the Rings trilogy. Where those shows might have a couple of fully computer animated characters in a scene,— Avatar — especially in its epic final battle — would feature hundreds, Avatar's virtual environments were much denser, as well. "To figure out how we were going to light and render these very big, complex shots," explained Letteri, "we went back to ground zero and questioned everything we had ever done. A lot of techniques that are used in computer graphics go back 20, 25 years. They were developed in the days when computers didn't have nearly the speed and power that they have today; and so there were a lot of shortcuts taken. Back then, you could never render something as big as what we were rendering for this show. Now that we could do that, what was the best way? Was it still appropriate to use these old ideas? We reexamined every technique we'd ever used. If it held up, great; if it didn't, we came up with something new."

This reexamination of techniques resulted in an all-new, more physically-based pipeline at Weta. "The goal of everyone over the years has been to make shots look real," stated Letteri. "But we decided to bite the bullet and make them real, to actually calculate the reality. So, for example, in the models for our main characters such as Jake, Neytiri and Tsu'Tey, we built their muscles as they really are in the body. In the past, CG character 'muscles' have been like balloons under the skin that move based on joint movement and so forth. What we did was simulate real muscles, fat and tissue."

The crew positioned Sam Worthington atop a banshee gimbal rig to capture dynamic rider motion. During the mocap session, a monitor displayed flight path animation, including right and left banks, dives and climbs. Baneham and Cameron stood facing the monitor, their arms outstretched, and replicated all the pre-set flight moves, signaling the gimbal operators to move the rig accordingly. / Cameron and Baneham captured flight path animation for both flying creatures and aircraft, manually moving quarter- and half-inch-scale models across the mocap stage. The data obtained provided Weta animators with brood-stroke information regarding flight direction and movement.

The new pipeline included a different approach to lighting, as well. In the past, Weta's lighting pipeline had been based on ambient occlusion, which would determine if something was in shadow or not, and calculate the light accordingly. Avatar called for a more streamlined lighting solution. "Traditionally," said CO supervisor Wayne Stables, "we would load all of our assets into a Maya scene, set up our lights and kick off our renders. We could get away with that when we were dealing with half a dozen elements in a scene. But on Avatar, we were going to be dealing with hundreds of thousands of CG elements. That made our traditional approach impractical. The Maya scenes would have become huge and impossible to navigate through."

The new technique, called 'spherical harmonics,' would not only determine if a subject was in or out of shadow, it would also determine the direction from which light was emanating. "For any given point on a surface," said digital effects supervisor Dan Lemmon. "it would tell us where the light hitting that surface was coming from, and light the object accordingly."

This physically-based lighting approach required more computation on the front end; but once those computations had been done, it streamlined the lighting process considerably. To light dense, intricate forest environments, Weta encoded harmonics onto its high-rez models, breaking them down into smaller, more manageable pieces. "Within any lighter's scene," said visual effects supervisor Eric Saindon, "all they had were simple bounding boxes of where the different trees were, the different ground covers and things like that, all broken down into small bits so that when we pushed that into RenderMan, it could handle it very easily."

Lighting was simplified even in epic shots, such as those in the final battle that featured hundreds of creatures and characters, along with huge numbers of aircraft, hardware and effects. "The destruction in those scenes is glorious and huge in scope," noted Weta visual effects supervisor Guy Williams, "but we were able to light them with only three lights. In simpler shots, we could put in a single light and get a very compelling, realistic result. We reduced the burden on the lighters by a significant degree, and at the same time we increased the reality of the scene because the lighting was much more accurate."

The improved lighting made for more believable Na'vi and avatar characters, as well — a major concern from the outset of the project since the CG characters often would he integrated with live-action photography, acting beside live-action characters. Just such a scene comes relatively early in the film, when Jake first awakens within his avatar body and, overcome with the joy of being able to walk again, breaks free of his restraints and runs out into the Hell's Gate exercise yard. "That scene is the introduction to the CG avatars," said Williams, "and it was a bit of a trial by fire because the avatar is in live-action plates with real human beings, and we get really close to him in the scene. So we really had to sell the reality of that character. As it turned out, it was beneficial for us to work on those shots early on, because it forced us to bring the reality of the CG characters up to that very high level."

The superconductor properties of unobtainium — the invaluable material being mined by the consortium — create the phenomena of massive rock mountains suspended over the Pandoran landscape. To visualize the floating mountains, Robert Stromberg and his team produced moving concept pieces, projecting artwork onto animated geometry. Weta Digital refined the look in its shots. Working from photographs of a mountain range in China and laser scans of real rocks, Weta modeled the massive boulders in 3D, layering displacement maps and textures onto the geometry. ILM later replicated the look, using 2D techniques, for its numerous shots of large-scale'e aerial combat.

Elevating the level of realism required that the lighters, as well as the shader and texture teams, work diligently to overcome the difficulties inherent in making blue skin look warm, alive and organic. "Blue skin is always going to look contrived," stated Williams. "Even if you create a blue-skin creature that is 100-percent correct, technically, it still isn't going to look real in a shot."

"You put humans in firelight," added Jim Cameron, "they look very natural. You put a blue person in firelight, and the orange of the fire cancels out the blue and you get a sickly gray. The first time we lit one of these characters with orange light, they looked terrible! It was like a zombie Na'vi. So Weta had to come up with a technique called 'spectral compensation,' where they would treat the orange light reflected by the blue skin differently than the orange light falling on the surroundings. They actually created a tract that said, 'That photon hitting the na'vi is coming back at a different spectrum than the photon hitting that piece of environment the Na'vi is sitting in front of.' It was all in the math, and it took a while to work out. Overall, we found that if we weren't in complete control of the lighting, these characters could easily wind up looking like blue plastic."

Artful texturing also helped to avoid a plastic toy look. Weta artists added discolorations in the blue skin, mimicking the uneven tones in human flesh. "A real person with peach-colored skin doesn't have even peach tones all over," Guy Williams said. "In a person's face alone, the coloration and the hues change dramatically from one area to the next — but your mind reads it as one tone. We figured the same thing would be true of blue skin. So we started painting more variation in the blue. The skin around the eyes was slightly darker; the cheeks were slightly less saturated; the skin around the lips was pinker. There were all these subtle tonal shirts in the skin. Then we started to put in blotching. That and the color shifts, even though they were very subtle, added a sense of reality to the skin."

Jake wins over a leonopteryx — larger and more fearsome than the banshees — and sides with the low-tech Na'vi resistance against Quaritch's high-tech military forces during an epic battle that concludes the film. As he had for the banshee, Yuri Bartoli pointed megapixel images of the leonopteryx's colorful markings in Photoshop, finishing the renderings to exquisite detail and then projecting them onto a 3D leonopteryx model in ZBrush to serve as a guide for the artists at Weta Digital. The leonopteryx was just one of hundreds of creatures featured in the film's climactic battle.

Courageous Weta crew members subjected themselves to extremely close-up photography of their faces, providing the team with reference for fine pore and other textural detail. Weta also had photographs of Sam Worthington and the other actors; but texture artists found that they had to exaggerate what they saw in those photographs — the freckles and other imperfections — for them to show up on the final blue-skinned avatar. "Our texture maps looked a lot more ratty than the real person's face." said Williams, "but by the time we put it in a render, there was a softening effect. So our end result looked real, even though we put a lot of over-the-top aging and weathering into the skin."

A new approach to subsurface transmission also helped to create natural-looking blue flesh. "Our new subsurface model is an absorption-based pipeline rather than a scattering-based pipeline," Williams explained. "It allowed for the proper transmission of blood colors coming through a surface. So even though we had this blue character, we could still get a lot of pink and orange transmission coming through, without altering the color of the skin. It made it more difficult to paint textures, but we reverse-engineered textures into our pipeline so our texture artists could continue working like they were used to."

Weta developed all of its lighting, texture and subsurface techniques on the shots of Jake's avatar awakening in the lab; and the photorealistic look of the character in that scene became the high bar Weta had to meet with all the Na'vi and avatars. "Jake had to work next to the live-action characters," Williams noted, "and all the other CG characters had to work next to Jake. After we got Jake's avatar to the point where we were really happy with his look, we put him up against Norm's avatar and realized just how far we had come. Norm instantly looked like a flat CG character in comparison. So we had to go back and do all the same texturing and other things to his character to get him to hold up next to Jake. Then, once we had all the avatars looking really good, suddenly we had problems with our CG environments. The plants around them in the jungle didn't look good enough anymore; so we had to upgrade the plants. Or we'd put their clothes on them, and suddenly the clothes didn't look real enough, and we'd have to go back and add many more polygons for the clothing. It was the same thing with the Samson helicopter. It looked gorgeous on the turntable; but when we put Jake inside the Samson, suddenly it looked like a plastic toy. So the Samson had to be up-rezzed. It was almost like a virus infecting everything. As we put our newest, best asset into a shot, everything else in the shot had to be promoted to that level."

Quaritch sends troops to shut down the remote Site 26, where Jake and Grace are using mobile link units to connect to their avatars, now working in alliance with the Na'vi against consortium forces bent on eradicating Pandora's native opposition. Live-action shots of the actors disembarking a Samson helicopter made use of the practical Samson prop built by Weta Workshop, with animated rotors added in post by Industrial Light & Magic. ILM, under visual effects supervisor John Knoll also added environmental elements surrounding the live-action.

When Weta had first signed on to Avatar, Joe Letteri had warned Jim Cameron and Jon Landau that it would take a full year to conduct all the requisite research and development and to get the new pipeline in place. That prediction had proved accurate; but the pipeline was more or less in place by the time Weta began receiving template files from the virtual production team in Los Angeles. The template's first stop was Weta's layout department, where animation technical supervisor Shawn Dunn would pull apart all the disparate elements and re-render them. "We'd line up all the original elements," Letteri explained, "replace what was in the template with our high-rez elements, and work out any problems we found. Sometimes things didn't translate well. On the stage they were doing things quick-and-dirty, out of necessity. So, for example, they would scale up a tree by a factor of 20 because they needed a big tree; but when we put in the high-rez model, it became a tree with really big leaves and an out-of-scale trunk. We had to sort through all of those kinds of problems." Weta also corrected character animation irregularities in the low-rez template. "We had to reconcile the footfalls of the actors to all the contact points on the terrain; or there might be characters going through each other in shots that had been done in multiple passes. There was a long layout process to make sure that everything was legitimate before it got passed into animation."

Dave Clayton, a lead animator on King Kong, had laid the groundwork for the animation effort at Weta, animating many of the motion studies of the creatures — such as the four-winged flight cycle for the banshee — in the early days of the production. Clayton continued to supervise a team of 45 animators when Andy Jones came on the show as animation director. "Dave was invaluable," Jones observed. "On this show, more than any other, my time was limited because I was constantly in reviews, trying to chase down stuff for final. While I did that, Dave kept the animators on track." Specific sequences were assigned to six animation leads — Jan Philip Cramer, Dan Barrett, Mike Cozens. Robyn Luckham. Alex Burt and Ben Forster — while Nick Craven animated previz for large set pieces. These leads and their crews would produce 110 minutes of computer animation — some 1,800 shots, many of which featured more than 1,000 digital assets.

The practical Samson was suspended from a crane over a minimal greenscreen set that consisted of foliage and ground cover. To build the large helicopter. Weta Workshop first constructed a smaller five-foot-long version, based on production artwork. When Cameron approved that prototype. Workshop artists re-created it as a digital model and then constructed the full-size version primarily out of fiberglass. Simulcam proved a boon to such greenscreen-heavy scenes, as its real-time compositing enabled Cameron to see the imagery that ultimately would fill the matte area, and frame his shots accordingly.

By far the most delicate and painstaking animation on Weta's slate was the many facial closeups, requiring animators to exactly match the actors' performances as photographed and captured on stage. "Perfecting those intimate moments was key to the film's working on an emotional level," explained Richard Baneham, who worked closely with Andy Jones and Dave Clayton throughout. "Subtext and the inner monologue really come from the facial performance. As an audience, when we see a piece of film, we immediately lock onto the eyes of a character and try to determine what the character is feeling emotionally. We had to have all of that emotion there in our closeups."

Re-creating those emotive facial closeups required the employment of facial capture data, HD reference material and, inevitably, keyframe animation. "It is everybody's dream to capture a facial performance and have that work perfectly on a character," stated Andy Jones, "but that's never really the case. Where the animator becomes important is in being able to recognize and add the details that are missing from the facial capture — a quiver of a lip, a slight movement of one eyebrow. It is that last ten percent that makes it real. We weren't changing what was there to begin with; we were just trying to get every last nuance out of that original performance."

While the single-camera head rig had supplied the requisite raw data for facial capture, it was of little use as reference for the animators because the camera — situated so close to the actor's face — resulted in imagery with a fish-eye-lens look. It also provided little information as to the movement of the upper part of the head, since that part of the actor's face sloped away from camera. As a result, the animators had to rely very heavily on the HD witness camera footage that had been shot both on stage and in separate facial animation capture sessions. "We put the actors in front of six or eight HD cameras," said Jim Cameron, "and I sat with them as we went through the script, scene by scene. We had them act out their parts for the whole movie, with full range of expression and emotion. It gave the animators an advance look at what they'd be doing in the facial animation."

ILM supplied background imagery for a shot in which Quaritch directs the battle from his Dragon gunship, adding support aircraft outside, plus hardware debris and dynamic battle explosions created through new simulation and rendering techniques that produced ultra-realistic explosions interacting with animated CG objects. Working in Photoshop, Stan Winston Studio artists designed a scar makeup for Stephen Lang, suggesting that the commander had survived the swipe of a very large predators paw. SWS fabricated the scar prosthetics and sent them to New Zealand, where they were applied to the actor throughout the shoot. Quaritch's gunship opens fire on the Na'vi. As they had for other aircraft in the show, Weta animators conducted a series of motion tests to develop flight animation for the Dragon, conceived as a massive flying fortress equipped with powerful weaponry. When Weta's burgeoning workload required that some visual effects shots he farmed out to other vendors in postproduction, ILM took on a portion of the final battle featuring the Dragon and other ships. ILM started with flight path motion capture and the Weta motion tests, and then finalized aircraft animation through keyframe techniques.

"We lived or died by our HD reference," Baneham added. "It not only guided the animators, it helped Jim in the review process: he could compare the animation to that original closeup reference, and say, 'Yes, that's it,' or 'No, that's not it.' Later in the process, as he gained confidence in what the animators were doing, he relied on that reference less. He'd sit and watch the animated scene without looking at the reference, and react to it on a gut level. We were all less reliant on the reference at the end; but the whole front end of the process was a marriage between the face cam, the reference cameras, the incredible rigging department, and Andy Jones, Dave Clayton and myself going through iteration after iteration of facial animation until it finally evoked the right emotion and achieved all the right shapes."

To realize the CG character faces, Weta built high-resolution facial models from scans of character maquettes, making some alterations to those original designs to ensure actor expressions would translate to the alien Na'vi and avatar faces. "People ask me, 'When do you think we'll be able to do humans?'" Jim Cameron remarked. "And my answer is, 'Sooner than we'll be able to do blue people with lion noses and pointy teeth and big eyes.' It would have been much easier if we were literally re-creating Sam Worthington, because the CG character would have mapped to the actor exactly, one-to-one. But, of course, why would you ever need to do that? Just shoot Sam Worthington! What we were doing was much harder. We were trying to get Neytiri's performance to match Zoe's, and Jake's performance to match Sam's, while also accounting for the physiological differences between them."

In the original concepts, those differences had been more acute. The Na'vi and avatars had featured a mouth that was slightly pulled forward, forming the suggestion of a muzzle, for example, and the eyes had been positioned more to the sides of the face. But when Weta animators conducted early expression tests to determine if they could match the actors' expressions with those initial models, they discovered that there was too much of a discrepancy between the actor's face and that of the character. The eyes, in particular, were key to matching the actors' expressions; and so, as modelers resculpted the original scan data in ZBrush and Maya, they pulled the characters' eyes more toward the front of the face, like a human's, and also incorporated the actors' lid folds and brow shapes.

Long before the final facial models were approved, Weta's animation team had begun building extraordinarily robust facial rigs for each of the characters — rigs sophisticated enough to compensate for the relatively low level of data that had been acquired through the production-friendly facial capture setup on stage. Andy Jones and Weta facial lead Jeff Unay developed muscle-based facial rigs using the Facial Action Coding System developed by Paul Ekman, a psychologist who has spent his career studying and quantifying universal facial expressions. "FACS allows you to encode all of the facial muscles numerically," said Jones. "If you want an anger pose, you put in the right numbers, and the system knows which muscles to pull."

Even with the help of the FAC system, however, rigging the characters for facial performance was the single most challenging aspect of the show for Weta's animators, and perfecting those rigs was a slow, laborious process. "The first time they showed me a piece of animation, it sucked," Jim Cameron admitted. "We were so deep into this thing, and we'd spent so much money, and we'd been through so many gyrations already— and then I see the first animation of Jake and Neytiri, and oh my God. I wanted to kill myself. But there was no pulling out of this; and so we played the hand out, and it got better and better."

Ultimately, Weta's animators would more than meet the facial animation challenge; but the true magnitude of the accomplishment would not be measured until their delivery of one of the most emotional scenes in the movie — the scene in which Jake confesses to Neytiri that he has infiltrated her tribe on assignment by Quaritch. Betrayal, rage and hurt all play on Neytiri's face, just as they had on Zoe Saldana's the night the scene was shot. "When we shot that with the actors on stage," recalled Richard Baneham, "there was a palpable energy in the room. Everyone was really there, in the moment, despite the mocap suits, the lights, the cameras. It was a very emotional scene for Zoe, in particular, and she just killed it. She and Sam had such a connection that night, you could literally feel the sparks in the air. Jim and 1 talked about the scene afterwards, and we both realized that we had a very high bar to meet in the animation, because it had to be at least as good as what the actors had done that night. Magic had happened, and we had to do our damnedest not to lose any of that magic, any of that emotional content." The scene went through many iterations and reviews with Baneham in the months that followed. "We didn't even show it to Jim during these intermediary stages. Only when we'd really cracked it did we finally show it to him — and his response was incredibly positive. There was a real sense of relief in the room. With this scene we saw that, yes, we would be able to get where we needed to, emotionally. The Weta animators absolutely knocked it out of the park."

Na'vi warriors, on direhorses, ride into battle against Quaritch's ground forces. Early motion study tests for the direhorse end other six-legged creatures had revealed that a natural-looking gallop could be achieved by coupling the front four legs, but offsetting the movement of the first pair by 10 or 15 degrees. While the extra two legs tended to blend with the other four In gallop and walk cycles, they were more evident when the creature was standing still, requiring Weta animators to devise a range of stomping and shifting behaviors for the extraneous limbs.

Production motion-captured real horses on the stage at Playa Vista, gathering Performance data that was then refined and modified for the CG direhorses by the animators at Weta. / Neytiri, in war paint, prepares for the final battle. Among the artistic challenges in designing the character was making her look sufficiently alien and yet familiar and appealing enough to make Jake's attraction to her natural and convincing. Neytiri and other pure Na'vi characters wound up resembling the actors portraying them, to greater or lesser degrees, because shared facial features aided in the retention and re-creation of original actor performances.

In the end, the emotional performances of the CG characters in Avatar must be attributed to the actors, most of all, as the performances are theirs. But credit must also go to Weta's ability to re-create those performances, down to the smallest detail, via the sophistication of its facial rigs and the talents of its animators. "The facial rig was absolutely one of the huge technical breakthroughs on this movie," Baneham asserted. "It helped to maintain consistency of character, down to the smallest detail and idiosyncrasy, as the animation went through many, many hands; and it took out some of the fundamental leg work. But the fine nuance of the actor's performance still had to be done by keyframe. The facial rig just freed the animators to concentrate on those all-important details."

"I think the facial performances are the biggest accomplishment of this film," Cameron noted, "and a lot of that was due to the character rigs. The final performance was maybe 10-percent keyframe, whereas it was probably 60-percent or 80-percent with Gollum. So we came a long way with the rigs; but that doesn't in any way minimize what the animators did. Keyframe not only gave us the fine nuances of expression, it also gave us the movement of the ears and the tails, which added a lot to the character's performance. When Neytiri is furious, we don't see only what Zoe did — we also see her lips pull back and her teeth clench; we see her ears flatten out and her tail switch back and forth like a lion's. The animators did all of that. They gave us 100 percent of what Zoe did, with a turbocharger on top of it."

"It had always been our intention to preserve the actors' performances," Jon Landau added, "and the talents of Andy Jones, Richie Baneham and all the animators in-house and at Weta enabled us to do that — absolutely. It would have been impossible without them."

Facial closeups and intimate, quiet scenes between characters stood in stark contrast to the large-scale, epic chaos of the film's climactic fight between the Na'vi tribes and the consortium's military force — what Cameron had envisioned as 'the mother of all battles.' But even there, amidst the dynamic action, between the skirmishes in the air and on the ground, Cameron incorporated character-centered moments. "You have to know and care about the characters to give a rat's ass about them once they are in the middle of a big battle," commented Cameron. "Even in the middle of that battle, there had to be intimate contact between the key players, little eye-of-the-hurricane moments. We needed to know who the main characters were, where they were in the action, and why it was happening. What were the bad guys trying to do? What were the good guys trying to do? The first cut of the battle was 37 minutes: but, applying all those philosophical precepts, I started focusing the action and throwing stuff out."

Even in its trimmed form, the final battle sequence runs more than 20 minutes and features thousands of digital elements — characters, environments, aircraft, vehicles, weapons and other hardware. Prominent in the battle are the small Scorpion and larger, Huey-like Samson helicopters. Weta animators did early tests of Scorpion and Samson take-off and flight animation, as well as motion studies for the Dragon, Quaritch's massive gunship. "It has four rotor systems that allow it to take off vertically, then fly off with great speed." said Andy Jones. "It's like a flying fortress with weaponry. We also animated the huge shuttles that bring crew and equipment to Hell's Gate."

Creatures would also figure prominently in the final battle, just as they do in earlier scenes. Prior to Andy Jones' arrival at Weta, Dave Clayton and a small animation team had conducted motion studies for some of the six-legged creatures, elaborating on the initial studies animated by Paul Kavanagh. "In these motion study tests," recalled Jim Cameron, "we discovered that the best way to create a gallop or run was to couple the front two pairs of legs and have them act like the single pair does on a dog or a horse — except one set of limbs would be 10 or 15 degrees out of phase, landing a little bit ahead of the next. And then, for a slow walk, we could have them move like an insect. We called that the ant walk, where every pair opposed the adjacent pair. So the front pair would be in opposition to the middle pair, and the middle pair would be in opposition to the back pair."

While the extra two legs tended to blend with the other four in gallop and walk cycles, they were more evident when the creature was standing still. "The most fun situations for the animator were when a creature was just standing there, shifting its weight or stomping its feet," said Jones. "That's when we could come up with different actions for all four front legs, moving them in a way that made them unique to that creature."

Animators looked at nature footage as behavioral reference for all the key creatures. Footage of foxes, for example, served as reference for the viperwolf, which Cameron had described as part-monkey, part-dog. "In addition to the fox reference," stated Jones, "we looked at cheetahs and wolves. Because the viper-wolf has such a long, skinny body, we even looked at greyhounds to get a sense of their long strides. We mixed all of those together in the final animation. We had some great tests of them on the prowl, too, looking much more cat-like — but we never had a place to fit that in."

In an ILM composite, a live-action ground soldier observes as banshees and Scorpion helicopters engage in aerial combat As did the other visual effects vendors, ILM worked from templates provided by the production. Template shots, reflecting to the frame Cameron's specific vision for the finished shot, obviated the need for creating shot handles — extra frames before and after the final-cut shot — thus conserving on rendering time, but complicating the application of motion blur since blur is typically calculated from trajectory data derived from the next frame in sequence.

The thanator had movements that were more evocative of large cat behavior. To guide the animators in the sequence in which the thanator chases Jake's avatar through the forest, production supplied them with motion capture of a stunt performer — as Jake — running over and under obstacles placed on the stage. "Jim had ideas for what Jake would be doing," noted Jones, "and those actions were in the motion capture. But what the thanator was doing was left somewhat up to us. We built up thanator actions, like jumps and swipes at Jake. In these types of interactive chase scenes, we weren't completely tied to the human performer's timing. Jim had cut together a lot of action, and we could mix it up as needed." Cameron then went back and shot virtual cameras to Weta's final thanator animation.

Weta conceived a wide variety of behaviors for the banshees, especially for the scene in which Jake, as part of his Na'vi hunting training, climbs to the top of a mountain, the site of a banshee rookery. "Jim asked us to come up with some ideas of what the banshees might be doing in this setting," related Jones, "and do some motion studies based on those ideas. One of the animators, Jalil Sadool, created a motion of the banshee taking off in a very aggressive way, vertically, and then diving down into the canyon. Jim loved that, and he ended up using it in the movie." Banshee wing animation was determined by its batlike configuration — a leathery membrane stretched between ribbed structures — and was guided by flapping motions and flight paths recorded on the mocap stage.

Weta inserted all its animated characters into high-resolution, richly detailed environments that were based on the crude assets created by the virtual art department. "We got a big library of these very simple trees and plants from Jim," said Eric Saindon, "probably three- or four-hundred different plants and trees, just sketches on basic cards. And then we built high-rez geometry of each of those, plus some variations, for a total of about 1,500 different plants, ground cover and trees. Individual dressing for each of those resulted in a library of probably 2,000-3.000 plants and trees. When we got the template file from L.A., we would do a straight match, matching their sketched plant with our high-rez version." After matching what was in the template, Weta artists filled out environments as needed, sometimes painting plants and ground cover to areas that required additional dressing, in other cases growing a simulated jungle in Massive. "We could use Massive to plant seeds and run a simulation that would grow trees in a very natural way. For example, if some of the trees weren't getting enough light, the simulation would have them die. The simulation created ground cover and other plants, too. It gave us a realistic-looking jungle, with very little effort on our part."

In another ILM shot, Scorpions and Samsons fire on a Na'vi village below. For all of its shots, ILM had to first convert the production's MotionBuilder scene files to its Maya-based pipeline, and then replace all the low-rez elements in those scene files with high-rez assets, including Scorpion, Samson and other aircraft models. To create missile trails and tracer fire for the end battle, ILM developed a new system for generating particle-based smoke trails from projectiles, which allowed artists to tweak settings and tailor looks on a per-shot basis.

To improve render efficiency, Weta implemented a pruning technique for its environmental elements. "We did a lot of early testing with rendering a tree close-up," noted Saindon, "and then throwing away geometry as we pushed it away from camera, but maintaining the same silhouette. So a tree that was 5 million polygons close to camera would be only 500 polygons when it was pushed to the background — but it would fit the same shape. Whether rendered close-up or far away, it had to look like the same tree. Pruning was a way of doing level-of-detail, without having to actually build different levels of detail. It just did it automatically."

A major setting featured in some 200 of Weta's shots is the Tree of Souls, the massive living conduit to the spirits of all the Na'vi who have come before. In night scenes, the tree's foliage glows, resembling illuminated patterns on the Na'vi skin and the bioluminescent plant life that can be seen throughout Pandora. Bioluminescence is introduced in the scene in which Jake spends his first night alone in the jungle. "There were both creative and technical challenges in this scene," observed Joe Letteri. "Creatively, we had to figure out the best way to introduce the bioluminescence. Does it just suddenly appear? There was one concept where Jake's torch would go out and all the biolum would come on, instantly. But that looked like we were throwing a switch. Eventually, we came up with a way to introduce the biolum slowly and subtly throughout the earlier shots, without it being a distraction. If you went back and saw the movie a second time you'd realize that you were seeing this glow evolve as everything else was going on — you're just not aware of it until Jake pauses and looks around." The bioluminescence was inspired by underwater life Cameron had seen on his diving expeditions, and was also reminiscent of the glowing aliens and craft in The Abyss, Weta created the glowing look using different techniques, depending on the shot. "For more distant bioluminescence," said Wayne Stables, "we could just paint the patterns and project them onto the environment. When we got closer to it — such as in the Tree of Souls — the bioluminescence was revealed to be little bits of plant life; and so we had to actually build plant structures in 3D."

At the height of the battle, Quaritch survives the crash-landing of his gunship and resumes his attack on the Na'vi in an Ampsuit. Weta Digital realized CG versions of the Ampsuit for ambulatory shots, backing in shots of Stephen Lang in the practical Ampsuit cockpit. Quaritch in the Ampsuit ultimately goes one-on-one with Neytiri atop a thanator. To create a more natural dynamic, Cameron choreographed the fight using two Stunt performers — one representing the Ampsuit and the other representing the thanator — on the mocap stage, rather than rely entirely on keyframe animation.

Another fantastic, surreal element of the alien landscape are the giant floating mountains, held aloft in the skies above Pandora by the moon's powerful magnetic field. To visualize the floating mountain range, Rob Stromberg and his team had produced moving concept pieces, projecting their artwork onto animated geometry. "That helped to define a look that Jim was quite happy with," Dan Lemmon remarked. "Our task was to create the same thing at a larger, more photoreal level. We get quite close to the mountains in our shots — closer than they had in their tests — and so we had to put much more detail on them. And because we were going to have light moving across the mountains to suggest cloud movement and time changes, we couldn't just use projected paintings for everything."

Working from photographs of a mountain range in China and laser scans of real rocks, Weta modeled the massive boulders, using Mud box to paint detailed displacement maps. "It was too much geometry to paint efficiently in a traditional paint package," said Lemmon, "so we built a texturing pipeline using maps that were reasonably high resolution — 8K — and layered those textures onto the mountains and rocks. We had a base pattern, and then on top of that we put a dark moss and algae staining that we'd seen in the photo reference. We added another layer of rust stains, as if iron was leeching out of the rocks, and more small-scale vegetation."

Some floating mountain shots feature waterfalls that dissipate into clouds of mist. "We used a combination of 2D and 3D techniques for the waterfalls," said Lemmon, "depending on the shot. There's one sequence where they are flying on banshees and they actually go underneath a waterfall. We had to make 3D water for those shots, using Maya Fluids and Maya Particles along with custom software to make those simulations more realistic. Others were done using fluid simulations that were written here specifically for waterfalls."

In addition to creating Pandora's natural world, Guy Williams' team at Weta Digital created the entire Hell's Gate complex in CG. "Because we built all of Hell's Gate," noted Eric Saindon, "Jim could drop a camera in anywhere, and we could render it. It took a lot to set that up in the beginning, but it enabled us to get those shots out pretty quickly at the other end."

Cameron blocked out Ampsuit action in performance capture sessions with stuntman Kevin Dorman. That capture data drove a motion base on which the practical Ampsuit was mounted for closeup and mid-range live-action shots of Quaritch at the control. Cameron, with the 3D camera rig he developed with Vince Pace over the course of the past decade, films Stephen Lang in the practical Ampsuit cockpit, against greenscreen. Lang also performed energetic motion capture action for the Ampsuit scenes.

When Weta started work on Scene 53 at the beginning of the Avatar assignment, it had taken a full six months to turn around those 12 shots. As the team neared its final delivery date, three years later, they were turning shots around in a week. "Early on," said Saindon, "a render for the jungle in Scene 53 would take 24 hours a frame. But once we got going, we could render something that had even more detail in a couple of hours. That was just a matter of our learning curve, and the way we optimized our systems."

Despite Weta's efficient turnarounds, by the close of 2008 it was clear to everyone that Avatar had grown too big for one company to handle alone; and so, shortly after the first of the year 2009, Cameron and Landau divided approximately 600 non-character shots among a handful of additional vendors. Industrial Light & Magic, under the supervision of John Knoll, took on 180 of those, including additional shots of the floating mountains. "We had to work out a pipeline to get parity of look with what Weta was doing for the mountains," Knoll observed, "and also make whatever variations were needed. We got the base meshes for the rocks from Weta; but they were using a proprietary procedural approach for the trees and plants on the rocks, and they couldn't share that with us. So we had to come up with our own ways to texture them and populate them with foliage."

ILM took on shots in the final battle sequence, as well, focusing primarily on those featuring the Samsons, Scorpions, Quaritch's Dragon ship and the Valkyrie shuttle. For helicopter animation, ILM started with the flight path motion capture, then added keyframe touches. "Our first take was pretty much an exact match to what Lightstorm had provided us," said Knoll, "and the second step was to smooth that out a bit. As good as Jim is at manually puppeteering helicopter models, sometimes you could see the bounce of his walking in the vehicle motion. We had to smooth all that out and give the animation more realistic flight dynamics."

ILM created a memorable moment in the film, in which Quaritch's Dragon ship crashes and explodes in the course of the final battle. "We developed a pretty impressive synthetic explosion simulation and rendering technique that gave us the ability to create realistic explosions that tightly interact with animated CG objects," Knoll commented. Also on ILM's slate was what became, after several rough cuts, the opening shot of the film: a 30-second flyover of a jungle. "The audience doesn't know at this point if it is a rainforest on earth or on Pandora, so we deliberately avoided any kind of alien plant cues to indicate that this is not earth. We also did a cluster of shots at the beginning, when the shuttle first arrives on the planet. We did the first part of that, with the shuttle passing over a mine pit and landing on the Hell's Gate tarmac."

Framestore picked up the sequence as Jake exits the shuttle, part of the company's 70 Hell's Gate shots, both exteriors and interiors. "Framestore did a lot of digital set extensions," said visual effects producer Joyce Cox, "building 3D environments that included Pandora forest backgrounds and the Hell's Gate facility. They also added animated aircraft, vehicles and digital doubles." Production assigned Frantic Films 180 shots to augment Rick Carter's live-action operations center set, including the holo-table map system and CG backgrounds that would be matted into greenscreen windows. "Frantic also animated aircraft and vehicles moving on the tarmac outside, and created the graphics that you see on monitors inside the ops center."

Pixel Liberation Front supplied graphics elements for nearly 200 shots of the interior of aircraft and the Ampsuit displays, and Hybride in Montreal developed another set of graphics for link room sequences, and also extended link room sets for 114 shots. Hydraulx delivered 25 shots in which they extended crematoria and cryovault sets. "They added digital doubles to both the crematoria and cryovault scenes, too," Cox stated. "Buf did a tunnel transfer effect for the link unit scenes indicating the transfer of consciousness to the avatar body. Blur contributed four 4K space shots for views of the mothership en route to Pandora; and, finally, Lola and Look Effects completed a variety of fixes. It's a huge movie — all but 110 cuts are visual effects."

On November 6, 2009, each vendor, including Weta Digital delivered the last of its shots — about 2,500 visual effects shots, in total, all rendered in stereo. From its beginning in 2005 to its completion in 2009, the Avatar production had been an exercise in dogged determination and roll-up-your-sleeves ingenuity. It also had been a test of forbearance. "The one thing I've learned in all of this is that you have to be Zen," Cameron said. "From a director's standpoint, when you're confronted with something that's disappointing, don't just throw up your hands, and say, 'This isn't working.' You have to get in and work with these smart people you're dealing with. You have to try things, run tests, do some wedges, play with different lighting. Mostly, you have to talk — talk, talk, talk. And I found that by doing all of that, you can get there, you can solve every problem. We solved some very tough problems on this movie.

"The other thing this experience has shown me is that when you've got all these possibilities, when you can do any kind of action imaginable, it is easy to run amuck. If you're a very impulsive filmmaker and you get vectored around by new ideas too much, you'll go crazy in this environment. You have to be very disciplined. Applying rigor and discipline to the process has been the biggest challenge, in fact. Early on, we came up with the principle of denying ourselves infinite possibility — which sounds wrong. You'd think you'd want to embrace the infinite possibility; but you don't, because you'll never get there. Ever. We stood by the principle of making a creative decision in the moment, and never second-guessing it. And just by making that decision, we had eliminated possibility. Every single day was about eliminating possibility, in fact."

In eliminating possibility, Cameron and his co-innovators also embraced it, and created a film that represents a phenomenal technological achievement.

But for all of its high-tech bells and whistles, all of its firsts, all of its groundbreaking methodologies, Avatar, ultimately, is a creative achievement, an evocative, heartfelt story about real characters engaged in real relationships, struggling to make their way in a real world. "I think the movie will be celebrated more for its small moments than anything else," concluded Cameron, "for the way that it immerses you in this world. There are places where the movie moves very fast; but there are also places where it slows down so you can hear the bugs, and feel the foot contacts with the ground, and enjoy looking at this bioluminescent world. For me, science fiction fantasy is all about the seduction of reality, about creating a reality so detailed and textured, the audience can completely surrender itself to it. That's what I wanted to do with Avatar."

Avatar photographs copyright © 2009 by Twentieth Century Fox. All rights reserved. Production still photography by Mark Fellman. Additional photos courtesy of Legacy Effects. Special thanks to Jon Landau, Terri DePaolo, Nicole Pitesa, Judy Alley, Ri Streeter, Greg Grusby, Jim Charmatz, Chris Del Conte, Steph Bruning, Katharine Ratnoff, At Shier and Chrissy Quesada.