MPC on track with Fast Five

The latest incarnation in the Fast & Furious series, Justin Lin’s Fast Five, features a dynamic train heist sequence in which the main characters attempt to steal high performance cars from a speeding train. Shot practically in Arizona, the scene was augmented with visual effects by MPC in Vancouver. We talk to MPC’s VFX supe Guillaume Rocheron.

fxg: Can you give me an overview of MPC’s work for the train heist sequence?

Rocheron: We did just over 240 shots for that sequence. The work involved a CG train, CG train extensions, the bridge, digital doubles for stunt work, some face replacements and quite a bit of environment work to tweak the desert location. We added a whole canyon in there too for when the car and Vin Diesel and Paul Walker are forced to jump into a canyon. And there’s quite a few greenscreen comps of desert tiles behind the guys.

fxg: In some ways I’m surprised there are more than 200 shots in the heist, because it feels very natural, especially the camera movement and all the dust – what kind of planning and previs required?

Rocheron: The whole sequence was previs’d, and then shot almost entirely by the second unit in the desert in Arizona and I think they did an amazing job with the stunts and camera work. There was also a greenscreen shoot with the actors in Atlanta.

The pace of the sequence and the editing is really fast, so there were a lot more shots in the previs than ended up in the movie, but the key moments were blocked in there. It had to change a little bit when there was a big crash of the truck into the train that didn’t quite happen when it was supposed to happen. So the truck crashed into the train in one of the stunts and it was basically wrecked so the sequence had to be re-worked a little.

fxg: Can you tell me about the shoot?

Rocheron: Jessica Norman, a visual effects supervisor at MPC in London, supervised the shoot in Arizona, as I was still working on Sucker Punch. There was a real train car with one passenger car and three baggage cars. That’s what was used to execute all the stunts and the framing. Our task was to extend the train by adding passenger cars and the engine to make it look like a high speed train. And the vehicles were stunt dressed vehicles.

They shot with 35mm cameras, and then there were a lot of Canon 7Ds hooked up on different parts of the train and vehicles to do little inserts. The whole sequence is a mix of both materials. They were shooting between three and five camera set-ups at a time. Some of the stunts were really complicated and dangerous so they tried to get as much coverage as possible to avoid having to re-do the stunt twice.

When the truck crashed into the train, they couldn’t re-do that because the whole baggage car was completely destroyed by the impact. The collision you see in the movie is the actual real collision between the truck running at full speed and the train moving.

fxg: Let’s talk about some of the assets you had to build for the sequence – the train is obviously one – how did you approach that?

Rocheron: For the train, we LIDAR’d the passenger cars and the three baggage cars and did an extensive photo shoot at night. The train was extremely reflective and the desert was a very bright environment. Ideally, for a good texture shoot you want to avoid as much specular and reflections as possible. So we went there at night and used a couple of light set-ups, orienting them so that they wouldn’t shine too much on the metal.

So we used that as a basis for our modeling and texturing. Regarding the shading and the look dev, we took thousands of pictures of the train in light and shadow at different angles to see the amount of reflections. There were quite distinctive metals on the train – the bottom was more aluminium-like and there were the painted parts and the roof catching reflections differently. So we had to dial in the Fresnels, for example. It was a really difficult thing because in a lot of shots we had the real train and right next to it we’d be extending the train digitally. And the cameras were moving around a lot!

One thing we ran into was that as the shoot was going, the train was becoming dustier and dustier, which meant the reflections would change and they wouldn’t behave the same from one shot to another. So we had to make changes on a shot by shot basis depending on how the reflections were behaving.

fxg: The shots around the train also have a heap of dust being kicked up – how did you approach that when it had to be digital?

Rocheron: There was quite a lot of dust happening, especially when the cars break or change direction. It had to be a very heavy sandy dust. That was probably the effect that gave us the most problems. We had to cut back-to-back with some live action shots and be extremely detailed in full sunlight. We used some fluid sims in Maya Fluids and Flowline to get the movement and resolution as high as possible, and then we’d use some proprietary shaders to send that back into RenderMan to really add in detail.

Once the truck is stuck into the train after the crash, sometimes the stunt team and second unit team had a hard time keeping the truck trailing on the ground, and sometimes it wasn’t making a huge dust trail. So we had to add that in on quite a number of shots. You’d see shots where the dust was there on the practical plate and then you’d cut to the next shot and it had to be our CG dust to maintain the continuity.

fxg: There’s one particular shot of the train speeding towards the bridge from the POV of the train. Can you talk about that one?

Rocheron: That shot was a plate from a camera hooked up on the train, but you couldn’t actually see any of the train – it was just looking at the desert and you were moving forward. So what we did was match move the shot and then add a digital train. It was incredibly close to camera. Then we added a digital bridge as well, and and the canyon in the background.

fxg: What sells that shot, I think, is the quick whip up to the train, plus the camera shake and dust, and the canyon in the background.

Rocheron: Yeah, things like the depth of field were something we could really do through Nuke, and it was a combination of 3D dust and a couple of elements. We also made sure we had volumetric lights shining through the bridge, so you would get light pulsating through the lens to really sell the speed and danger.

fxg: So the train and truck then crash into the bridge – how did you do those shots?

Rocheron: That was interesting because there was no bridge. There’s one shot where you see the truck crashing into the bridge, so second unit set up charges onto the truck stuck in there and blew it up. When we added the bridge in there, we realised that the truck still had too much forward momentum because there was no actual impact to go with the explosions. So we had to augment that with CG fire to hide the collision points and then deform the bridge to sell the fact that it was still moving forward. We snapped some cables and had the main beams deforming – then we added CG fire and debris and fire, which managed to sell the impact.

fxg: Where did you use digital doubles and face replacements?

Rocheron: It was pretty much throughout the whole sequence. The main actors weren’t in Arizona for the second unit shoot, so it was all stunt drivers. On some of the stunts, like when the Corvette jumps out of the train, it was an empty car. So we had to add a digital Vin Diesel in there. Then when Vin is driving along the train to try and pick up Paul Walker’s character before he crashes into the bridge, it’s a stunt driver in the car so we had to do a full 3D face replacement.

There were also a number of shots where they’re fighting in the truck with a blow torch and there’s one shot where we had to do a face replacement shot on Paul Walker almost full screen. It was a very quick shot, but because of the lighting changes and strong light direction, it was tough. Also the head shape and hairline of the stunt guy compared to Paul Walker was so different that we couldn’t just use 2D elements to patch the head – we had to do a 3D roto of the face and render a full 3D head with facial expression and hair and skin shading and comp it on top of the stuntie.

Once the drop into the canyon, some of the shots are a mix between real stunt guys jumping off a cliff. We replaced the cliff they jumping off with a digital canyon, and there are some shots where we added digital doubles which made them full digital shots with a digital canyon and digital actors.

fxg: How did you acquire the body and faces of the actors?

Rocheron: It was all done with cyberscans and then we also did a hi-res texture shoot using polarized light to eliminate all the specular and reflections. We built that in Maya and used our shader library for skin and hair, and used Furtility for grooming say Paul Walker’s hair.

fxg: So did they really drive a car off into the water?

Rocheron: Yes they did! Basically there was this quarry in Atlanta. Our canyon at the end is about 300 feet high, but I think the location was a cliff was about 50 feet high. They basically threw an empty car in the water and they had two stunt guys jumping from the cliff edge into the water. We used parts of the live action to comp into the shots. Sometimes the car is digital and sometimes we roto’d out the stunt car so we could get the movement of the water interaction and we ended up replacing the canyon cliffs. It was a mix of practical elements and digital elements that we massaged to make the final shots.

fxg: You mentioned the greenscreen shoot with the actors – how were those plates brought together in compositing?

Rocheron: We had greenscreens shot in Atlanta and we had to put desert backgrounds behind the guys. With the greenscreens being shot after the desert plates, there was no way to predict which camera angles and camera moves would be used, so we shot tiles of the desert running at different speeds and various angles and we animated the stitches in Nuke to create animated environment bubbles. We would match move the shots, have a 3D camera in Nuke and the camera would move in the environment the same way that the camera in the greenscreen foreground plates would be moving. It was actually a very extensive use of Nuke’s 3D projection capabilities.

fxg: What were some of the tools you used for the overall work?

Rocheron: We used Maya for a core tool for everything that was 3D-based. We have our internal toolsets for handling geometry and volume rendering. Everything is rendered in RenderMan and all the compositing is done in Nuke. It was one of the first times we had completely switched to Nuke at MPC Vancouver.

It was really interesting work because everything had to be invisible effects. So even though the action was pretty crazy, a lot of it was filmed for real and the audience never should be feeling that there are visual effects in the film. It had to be as transparent as possible and that meant we weren’t really doing things that were over the top.