Several of the visual effects supervisors on Charlie and the Chocolate Factory tell Alain Bielik how they managed to incorporate complex CGI.
Tim Burtons fondness for practical effects is notorious. On all his movies, he has favored prosthetic make-up, miniatures or stop motion puppets to carry his vision, turning to digital tools only when no other option was available. When he set out to adapt Charlie and the Chocolate Factory for the big screen, he hoped that this approach would again suit the project. Roald Dahls beloved childrens story tells the tale of a young boy who wins a contest to visit, with four other children, the mysterious chocolate factory of Willy Wonka (Johnny Depp). The plant turns out to be a maze of magical rooms and machines manned by amazing little people, the Oompa-Lompas.
Although the fantasy world of Willy Wonka was ideally suited for Burtons unique visual style, it required the creation of a universe far more complex than any practical effects could achieve. When we started on the project, we estimated the visual effects shots count at 300 or 400, but it ultimately grew to 800, the majority of which were created by The Moving Picture Co. (MPC), notes overall visual effects supervisor Nick Davis (Harry Potter 1, Troy). It grew enormously, and one of the reasons was that Tim was initially a bit hesitant about using too many digital effects. He wanted to do as much as possible in camera and use CGI only when absolutely necessary. For example, one of the most ambitious sequences in the movie involved dozens of squirrels performing very complex actions. We had three options to create them: train real animals, build animatronics or use CG animation. We hoped that we could solely rely on the first two options to create most of the shots, but the movements that the script called for turned out to be too extreme to be accomplished in-camera. In the end, Framestore CFC produced more than 70 shots of very complex CG animation, with Jon Thum supervising the effort. It was extremely difficult, as some CG squirrels ended up full screen!
Another example of a sequence outgrowing the initial plan was the Oompas dance numbers. The little people are all played by one actor, four-foot tall Deep Roy. Originally, Davis intended to shoot Roy with a motion control camera and combine many passes to create a crowd of Oompas. But this approach met with unexpected difficulties. Every plate with Deep had to be scaled up by a factor of about 1.7 to reduce him down to two-and-a-half feet, the size of an Oompa. That meant that if the camera traveled, say, 20 feet in three seconds on the set, it had to cover 35 feet in the same amount of time when we shot Deep! There was only one motion control rig in England the Cyclops that was able to accomplish this feat. Even then, there were shots for which it was just impossible. When the camera had to go from four feet up to 18 feet on the set, it meant that the Deep Roy plates had to be shot by a motion control camera that could go up to 30 feet and there is no such rig So, the only way to integrate the Oompas in those big sweeping camera moves was to turn to CGI. Plus, at two-and-a-half feet tall, an Oompa doesnt fill a lot of space on screen. We realized that in order to compose interesting shots, we needed many more Oompas than first planned, which meant even more CGI! Thats how what was intended to be mostly a compositing job grew into an enormous CG animation project, with motion capture, facial capture, crowd, cloth and hair simulation
The Oompas shots were awarded to MPC, where more than 550 shots were created, including the Chocolate Room sequence, Violets transformation and several flashback sequences. Three different techniques were actually used to create the Oompas. The first one was filming animatronics scale replicas of Roy built by Neal Scanlan (Babe) for long shots of the characters sitting in a chair or in the boat. The second technique involved shooting Roy on greenscreen or bluescreen (depending on the color of his costume) and compositing him in the background plate. There was one dance number for which we had to create more than 50 Oompas, MPCs visual effects supervisor Chas Jarrett (Troy) recalls. We first imported the motion control camera move from the previsualisation. Then, we shot about 55 different passes on the real set with Deep changing position every time and repeating his dance routine with slight variations. For each pass, we needed between 10 and 15 takes, which meant that we ended up shooting 700 or 800 takes altogether. It went on for months! The various elements were later combined in Shake.
When neither animatronics, nor bluescreen worked, MPC turned to CG animation. The studio organized an extensive motion capture shoot with 24 Vicon MX 40 cameras gathering data from 70 markers placed on Roy. The actor performed the various dance routines, but also the daily activities of a typical Oompa. This data was either used directly to drive CG models in Maya, or transferred into MPCs proprietary crowd simulation system, ALICE, to create a library of movements. The facial motion capture was carried out at Eyetronics with an HD video camera mounted sideways, Jarrett continues. This set-up was designed to capture Deeps facial movements with the largest number of pixels possible. Deep was filmed lip-synching the songs while a fine grid was projected on his face. There were no tracking markers. Eyetronics did 24 cyberscans per second and conformed our polygon mesh onto their 24fps cyberscan meshes. The advantage was that it kept the volume of the face perfectly. Eyetronics provided us with one mesh that changed shape instead of one mesh per frame.
The cloth simulation was carried out in Syflex with an unusual methodology, according to Jarrett. The Oompas wear a one-piece costume, which means that any time they raise an arm, it pulls up the fabric on the opposite leg Every night, we ran 10 different cloth simulations that we applied to one character animation. In the morning, we reviewed the simulations and selected the appropriate one or modified the values to run ten new simulations. If we had 50 Oompas to create, we selected ten cloth simulations that were then randomly applied to the digital characters.
The Mother of All Rides
Besides the Oompas-Lompas, the other challenging aspect of the project for MPC was the creation of the white tunnel ride, a two-and-a-half minute fully CG sequence in which the characters embark on a surreal boat manned by 54 rowing Oompas. The boat travels on a Chocolate River into fantastic galleries. I knew it would be impossible to simulate the entire river in one go, Jarrett explains. So, we broke it up into sections, leaving fluid dynamics simulation for very specific elements. The Chocolate River was basically modeled as a flat surface, the flow being defined by animated textures created by complex shaders.
Fluid simulation was used solely to create the splashes of the liquid chocolate on the tunnel walls and columns. We took a CG element from the set, say a column, and imported it into Realflow. We then threw particles at it to generate all kind of splashes. The animation was saved in caches that could later be imported into Maya one by one. This meant that we were able to select perfect splashes for each set element. If one splash didnt work, we could just import another one. There was no need to run the whole simulation again. A complete fluid simulation of the chocolate river would have taken three days. With our approach, everything was pre-cached out.
In order to create effects like ripples and boat wakes, we used particles to create procedural displacements on surfaces. Each particle had a radius value, which grew with time. So, as the particle intersected a surface, this radius would expand and be rendered as a circular displacement. We used this when Realflow fluid simulation particles collided with our river surface, to create the effect that the Realflow particles had splashed into our river surface. We also used it to create the wake from the boat: we emitted lots of particles from the intersection-line between the boat and the river surface. This left a long trail of particles behind the boat. By rendering these particles as displacements (which all overlapped each other), we created the ripples behind the boat. Additional displacements were added using geometry displacements, a tool which takes arbitrary surfaces and seamlessly adds them together at render time.
All these elements were then integrated into one seamless CG animation by a new tool called PUDDLE and rendered in RenderMan. Tight shots on the boat were filmed with a full-size model mounted on a gimbal rig in front of a greenscreen. On the other hand, any shot in which the entire boat was visible was entirely computer-generated, including the characters.
Crafting a Glass Elevator
While MPC was busy creating the underground environment, Cinesite was tackling the interior and exterior shots of the factory itself. Half of the 160 shots were accomplished with models, the other half with computer animation. Jose Granell (Harry Potter) supervised the 10-week miniature photography of a massive 1/24-scale model of the factory and the town surrounding it, constructed by Cinesite Models. It was a major undertaking by any standard: 120 feet long, 80 feet large, 25 feet tall, 1,500 individual houses equipped with 12,000 windows, thousands of fiber optics lights Filmed in several passes, the model was digitally enhanced at Cinesite, where visual effects supervisor Sue Rowe oversaw the project: In order to populate the model of Wonkaville, we used Maya to build 3D vehicles and we added 2D elements of cats and chimney smoke. We also put in a lot of CG people using R.E.A.C.T., our proprietary crowd simulation software, which we had initially developed for King Arthur. We wrote a set of rules to guide their behavior: they should not walk into each other, nor into cars or walls; they should go faster here, slower there, etc. The animation was based on motion capture. We built a library of movements corresponding to daily life activities. For each activity, we motion captured about ten different cycles, one person at a time. The various elements were then composited in Shake.
The major part of Cinesites effort focused on the creation of the interior environments of the factory with the glass elevator sequence as the center point. In this sequence, Wonka takes his guest into a magical elevator that has the ability to move in any direction. Depp and the other actors were shot in a glass elevator set piece on a bluescreen stage. We had a lot of blue spill and unwanted reflections to clean up, but it was worth the effort, Rowe comments. It would have been much more difficult to create the actors many reflections and refractions in the glass, had we decided to shoot without the set piece. The elevator was the only real element in the shots; the rest was entirely built in the computer. About half of the CG environments textures were derived from photo references; the other half was directly created in Maya (associated to RenderMan). Surprisingly, very little of these rooms had been fully designed when we started on the show. It gave us a fantastic opportunity to participate in the development of the look of the film with production designer Alex McDowell.
Although most of the shots feature the real actors, many shots required a totally digital approach: Tim Burton wanted the elevator to zip at extreme speed, which was impossible to achieve with the motion control camera. We soon realised we needed to create believable digital doubles of the principals, as the real actors could not be moved at such speed. We used cloth simulations on Willy Wonkas coat and hair to make the CG characters more dynamic. For shots of the elevator flying outside, above the factory, Cinesite used reference photographs of the miniature to create textures that were projected onto 3D geometries. This was combined with a matte painting by David Early to extend the model sets beyond the scope of the original designs.
Make the Surreal Look Real
For all involved, the hardest aspect of the project was its very stylized approach. The factory world is so surreal in design and in concept that it was very difficult to ground it in reality, Davis concludes. It was one of the hardest projects I have ever worked on. When we went in, we had no idea of how hard it was going to be. But this surreal world also provided unique opportunities to craft visual effects unlike anything that had ever been done. Working on a project of this caliber and with Tim Burton was kind of a dream for all of us, Jarrett adds. Actually, when we started working on the show, a lot of people at other facilities from all over the world called us to see if they could join us. It was that exciting!
Alain Bielik is the founder and special effects editor of renowned effects magazine SFX, published in France since 1991. He also contributes to various French publications and occasionally to Cinéfex. He recently organized a major special effects exhibition that opened Feb. 20 at the Musée International de la Miniature in Lyon, France. Displays include original models and creatures from 2010 Odyssey Two, Independence Day, Ghostbusters, Cliffhanger, Alien Vs. Predator, Alien 3, Pitch Black and many more. The exhibition runs through Aug. 31.