Rendering a Fast and Furious CG Train

Rendering a Fast and Furious CG Train

Dust and sunlight created all sorts of reflection problems in pulling off the train. Courtesy of Universal

Suddenly, CG trains are all the rage, what with Unstoppable, Source Code, Water for Elephants and now Fast Five.In fact, for the thrilling train heist that opens the fifth Fast and Furious film, MPC Vancouver (under the supervision of Guillaume Rocheron) handled around 240 shots. These included a CG train, CG bridge, CG dust, extension to a practical train, digital doubles for Vin Diesel and Paul Walker and other digital stunts and face replacements and CG sets for the entire canyon shot nearly all by second unit in the flat Arizona desert.
Of course, there were quite a few challenges in mixing and matching practical and CG elements (since Rocheron was working on Sucker Punch, MPC's Jessica Norman from the London office was the onset supervisor in Arizona). But the primary focus was the train. While there was a real train on set with one passenger car and three baggage cars on set, MPC's job was to give the illusion of a high-speed train, which ultimately crashes into the bridge. That could be accomplished through extension by adding an extra car and the engine, or by replacing with a full-CG train, depending on the shot.
"For that train, we took thousands of pictures from different angles and different times of day to really identify how the reflections were behaving, which presented a challenge because we're in a very bright environment," Rocheron explains. "Normally, we try to avoid as many reflections as possible. You want the most flat-lit train, which wasn't doable, and because of the size we couldn't do a cross polarized shoot like we do for actors. So we had to create some artificial lighting to be able texture very flat lighting references so none of the textures would be biased with already baked in reflections or tints."

The bridge was patched together from different reference that fit with the setting and circumstance.

However, dust became a problem, too, because that altered the reflections in every shot as well.
"And the train was made of different types of metal, so it was going on a shot for shot basis," Rocheron continues. "You have the bottom section that's made of aluminum, the center section that is a glossy metal and then the roof that is made of a different metal. We had a hard time [getting] the right values for shaders to really make sure that the reflections and the Fresnels were working correctly. Obviously, the reflections changed dramatically depending on the time of day. We're rendering everything in RenderMan and we have our core shader library, so every metal shader is made of in-house shading components with different specular models and others."
Meanwhile, the digital doubles posed an interesting challenge because the sequence is a mixture of second unit photography in Arizona with stunt doubles. And for shots where you see the actors in close-up, it was shot on a greenscreen in Atlanta with a partial set train.

"So we either had to do a face replacement for every stunt done in the desert or add the digital actors in the vehicles," Rocheron suggests. "Originally, we thought we could get away with 2D face replacement solutions, but very quickly we realized that some of the shapes of the stunt double's heads were very different or the hairlines didn't match. And because the vehicles are moving a lot into the environment, the lighting changes quite a bit.

The practical cliff was replaced by one that looked more like a desert canyon, and CG doubles.

"So we decided to build full-on 3D digital doubles and roto'd in all the actors and vehicles and put our digital actors in there and we traced the head with a 3D version of it. There was even one shot where we had great second unit footage of a fight with Paul Walker on the truck, and they really wanted to keep the shot so we had to replace the stunt double's face, which filled two-thirds of the screen.
"The other challenge was to marry the greenscreen photography with the desert photography. What production did was shoot tialed backgrounds in the desert with three different cameras at different speeds, at different angles, on different locations, at different distances from the trucks. We could then stitch that footage and create some environment bubbles in Nuke, which we could basically apply on the greenscreen footage. We would get forced perspective and dust from the desert and we really could match the environments as seamlessly as possible."
And, of course, when Diesel and Walker do the jump in the canyon with the Corvette, it's a mix of real stunt work and CG replacement and enhancement. "The stunt was used actually for having a practical hit in the water and practical splashes with the car and the guys," he adds. "We took off the practical cliff that was 50-feet-high and replaced it with one that was 300-feet and it looked more like a desert type of canyon.
"There's a bridge in the sequence, which is also digital. Nobody would really build a bridge like this with those sheets of concrete footing, but we used different references that we patched together that work well for the section that fits onto the desert as well as the section that fits above the canyon itself.