Can you give us an overview of Digital Domains work on 2012?

David Stephens: The bulk of the work picks up basically about half way through the Escape from LA sequence. We actually pick up with a nice shot flying over the Golf course which is at the base of the Santa Monica airport. Which you see a big crack open up and swallow several houses and trees. Then it picks up basically at the airport. There is actually a small number of shots where the family is sitting there on the tarmac. They had a moving set that had the ground actually able to flex and wobble. We had to extend that off in to the horizon. Add on all the planes and other props that were on the runway's reaction to that, so they had to wobble the buildings. Of course, structures and hangers would be falling down.

From there that leads into the actually collapse of the runway. The enormous fissure that opens up, and the flight through that and they emerge of course downtown and fly through it. It eventually ends up with the last two shots of seeing California sliding off into the ocean. That chunk defines the majority of what we did. We did do a little in Washington DC, with some ash shots, and the Washington Monument collapsing. The vast majority of everything was in the latter half of the Los Angeles sequence.

What was the biggest challenge on 2012?

Stephens: Mostly data load, you're talking about a tremendous number of buildings and structures, large scale land scapes and what not that were crumbling. Most of this was an exercise in trying to apply rigid body type techniques and various things. Very, very large scale scenes, large numbers of buildings a lot of complexity in the building or structures. Having that tied in with a vast array of volumetric effects also that had to cover these very large areas. Lots of planning and logistics in terms of trying to put together so many things that had to be so many different layers. Of course your later layers had to depend on the previous ones, it's just that whole thing of trying to organize that and keep it straight, it was a lot of work.

How do you go about creating the destruction in Los Angeles scene?

Stephens: When we started off we had a lot of extensive previs done on the shots that we received,. At that point we had allot of choreography they had actually planned out for when what building tips into what building and various things. The first part of it would be just going through and blocking and make sure we had gotten the director's vision of the action going on in the scene.

You have to start off usually with the primary the largest structure or objects that we were having to break up. This would have to go through the demolition pipe line which the buildings themselves were all most of them were picked out of real buildings from downtown and others were things that had been modified or changed a bit. You had to have the buildings built to spec, so that it would work with the destruction chopping gizmo, that would actually break the thing up into small pieces. It would then go through an effects rigging process to have all the joint strengths and constraints that glues the building back together so that it essentially remains a cohesive whole.

Then you would have to apply whatever previs animation that had been done. That would have to be applied so that if the building is actually falling over. That initially was supplied by the animation we were doing, then selectively the parts of the building were released and turned dynamic and those things actually crumble and look pretty. At that point it's really getting those performances bought off on and getting those looking nice. Then if other simulations around them whether it's like small debris and glass or the dust and volumetrics. All those would be starting on those buildings or if that building crashed into another one then in allot of cases you would sim that second building after the fact using the first one as a collision body. You just keep stacking these things and working down stream until you have built up all the layers necessary.

How did Digital Domain incorporate Bullet (RBD system) into their pipeline for 2012?

Stephens: Yeah, Bullet is actually a good part of the pipe. Essentially, I think this is becoming more common is taking the game engine physics and being able to bring that in to one of the large packages. In this case it was Houdini, but we essentially wrapped up Bullet into a plugin in Houdini and built a pipeline around that. That eventually ends up being everything from a lot of the things that actually chopped up the geometry and glue them back together, joints and constraints. All that is then set up to work with basically help bullet conceptualize those dynamics, joints and constraints between objects. You basically end up with a series of plugins or networks that support the basic Bullet engine at that point, that was a big part of our process.

Ryo Sakaguchi: When we first got on the project, the tools that we had in hand, did not handle this amount of destruction. In the beginning there was a phase for about three months that we were unsure if we could do this size of destruction. We ended using the Bullet route, but even when we were developing it obviously we were doing a lot simulation tests, it took us three months to even tell ourselves that "oh maybe we can do this project" with the route we took. There was definitely a phase when we were we scared of whether we could really pull this thing off or not.

At what point did you know that you could pull off the required effects for 2012?

Sakaguchi: There is usually one test that we run. Which we had run which was in the third month, it was a small building test where it fell over and it looked really realistic when it fell down. It fell down on the ground and it squashed in half, those are the moments we were like maybe you can this project after all.

How do you know where to put the most work into a shot?

Sakaguchi: We kinda know usually when it is far away we can get away with a lot of hacks and tricks. If you go right up on the breaking building we know it's going to be tougher, so we tackled the close up ones first to make sure we can handle those close ups.

What was your favorite sequence to work on?

Sakaguchi: There is one shot were the plane is flying through, the camera pans, and it kinda goes with the plane. You basically so the whole Miracle Mile collapsing, starting form the Japanese building, the Samsung building over here and the Graves church over here, pretty much the whole street of Miracle Mile collapsing. That would be my favorite shot, for the sheer amount of destruction we got to show off.

How were you able to create such realistic effects for the earthquake sequence?

Sakaguchi: I think it's the complexity of the destruction and we achieved that by considering physical building structure within the model we were building. So we ran allot of test to try to collapse this bungalow model, to get that to realistic we first start with chop the building up and let it collapse. That looked kinda okay but it did not look that realistic. What we ended up doing was we made all the pillars with in that structure, pretty close to what the physical architecture would do. We had the pillars running vertically on the wall and we and the roof top holding pillars, these were all modeled to do precisely to what the real thing would do. We would even set constraints on the wall to be a little bit stronger or the supporting structure to be stronger. Then you run a simulation and the roof would collapse fast because it's built that way. The strength is weaker and you see some parts of the house are go first but you can tell the supporting structure is trying to hold on to it. That kind of detail in how we modeled it, constrained it and rigged the dynamics really added to the complexity of the simulation.

Can you tell us about how your team was able to pull off the effect of the earth splitting during the wide shots of the escape?

Marten Larsson: We had a lot of what we called the fissure. We were concerned with how far you could see in the fissure, and how to generate that for so many shots. You normally see a few miles down the fissure so we actually came up with a little library system. Where we basically took a chunk of the ravine or the fissure and pre-simulated it. We pre-simulated dust for it and even a few things that could fall over. Then we generated a few of these blocks and repeated the blocks with a little bit of offset and timing down the fissure. That way it worked out pretty easily and fast so we could dress up the first pass of the fissure. We could just draw path and it would put these blocks down. We started up with random offsets on each block and pretty fast we got a first test of what the shot would look like and hand adjust timing or swap out blocks. That's kinda how we avoided redoing the fissure in every shot.

Initially, our plan was for each of these blocks we would simulate the top surfaces as well as the road, building and cars. In the end we realized that it's pretty forgiving with the dirt collapsing but you could quickly see repetitive patterns or the director wanted specific things to fall down and add an extra house here. It ended up being more work than it was worth to actually pre-simulate all that. After we had the fissure we started hand placing things we were gonna throw over the fissure. They did not interact much with the dirt because they were free falling. We could simulate them separately.