SIGGRAPH: All About 'Avatar'
I recall watching the 2010 Golden Globe Awards a few months ago and seeing my childhood hero Arnold Schwarzenegger up on stage saying something like:
"Yar dis Abadah is dee greathest moofee. Ya ya."
A great truth then dawned on me -- when you have Arnold's attention, you're onto something big. And that is just what Avatar is. Big. With over 2000 visual effects shots, and a team of 1000 different artists working together, what you end up with is a production of epic proportions. Scenes are brimming with polygons.
I attended a talk on Avatar by the artists at Weta Digital (Dejon Momcilovic, Kevin Smith, Antoine Bouthois and Peter Hillman) at Siggraph 2010 and the amount of painstaking detail and hard work that went into each tree and cloud of Avatar's alien planet Pandora become blatantly obvious. It was hard work just keeping up with them. I tried my best.
Most of the problems they encountered with Avatar dealt with the film's massive scale. Setting up the lighting for a city street is one thing but how about setting up the lighting for a whole planet? And having it be consistent across each shot in the film? Kevin Smith talked about some of the techniques they used to keep things simpler. One idea was to build a light "rig" for all of the trees and vegetation. This created shadow maps for each of the plants at varying angles. When a scene was eventually lit, the plants could then pull up the specific shadow map which closest matched the scene light source. This saved the time of having to calculate shadows since they were all already baked in and ready to be loaded up.
Another topic Smith touched on was Image Based Lighting. This technique eliminated the need for traditional 3D point lights in exchange for something faster to set up and closer to real photography. The team used various environment projections to light scenes, usually in conjunction with each other. In one example, the hero character Jake was flying on one of the winged Banshee creatures. A basic jungle environment was dropped in to create the basic ambient light. An additional blue fill light environment (much like you would see in a photography studio) was laid on top to give more kick to the blue skin of Jake. And finally another environment of bright sunlight was added create a rim light around the character. With so much real world information in each of the environment maps, the lighting created was almost immediately photorealistic.
Antoine Bouthors and Peter Hillman both talked about an interesting new technique called "Deep Compositing". Instead of rendering regular alpha mattes and beauty passes with geometry "holdouts" or "cutouts", a separate "Deep Alpha" pass was created which contains information about where each pixel is in 3D space for each frame. So even when you have insane camera moves and objects overlaying other objects and subsequently vice versa, the "Deep Alpha" pass is able to position each pixel of your passes in it's correct Z depth. This helped save render time because elements could be rendered independently of each other without having to worry about other elements overlapping them. It was all taken care of with the Deep Compositing tool.
Both authors also mentioned a few things about volume lighting effects in the film. Since the film was going to be stereoscopic, none of the volumic elements-- such as clouds, muzzle flashes, and God rays -- could be 2D mattes or footage. They had to work in 3D space in order for the 3D to be real, correct 3D. A tool that Weta developed was a cloud program that could model and customize clouds for each scene extremely quickly with meta balls and noise algorithms. Shadow maps, like the ones mentioned earlier for the plants, were similarly used for the clouds in each scene, so that lighting could quickly be changed depending on what the director wanted. "Deep Compositing" also became a useful tool for the volume renders, as they would not need to be re-rendered if there were changes to the characters overlapping the clouds or God rays. The "Deep Alpha" channel enabled the volume effects to exist independently. This helped the team immensely in simplifying their workflow and avoiding re-rendering elements.
There were a few more tidbits of information, but the main theme running throughout the talk was really about early preparation and attempting to keep things simple. The creation of Pandora was such a huge task in and of itself that the team at Weta really wanted to anticipate and avoid any messes or nightmares that would come up later down the pipeline, especially with such a huge army of artists working on the film. Any extra steps or complications that could be prevented were worked out and resulted in much more efficient use of their time and energy. They worked hard but they also worked smart.
I want to thank Weta Digital for giving such a thought provoking talk on Avatar.