How They Did It: Live Free or Die Hard
Modern VFX For an Old-School Stunt Flick
Elsewhere, more apparently ordinary shots demanded the creation of close-up, photoreal cars. In Live Free or Die Hard, the visual effects work is often both more and less than meets the eye.
The mandate was to bring a certain physical reality to the production. Director Len Wiseman – on his third film a VFX veteran, with Underworld and Underworld: Evolution already under his belt – wanted to set his action apart from the more CG-heavy competition. “Len was insisting on the fact that, because we’ve got Transformers and other big CG movies coming out, this one has to feel more real,” says Chris Del Conte, VFX producer at Digital Dimension. “It has to be embedded in some kind of practical reality in order to give it that edge of being a Die Hard.”
Other companies – including The Orphanage. R!ot, Pixel Magic, and Amalgamated Pixels – did VFX for the film, but Digital Dimension got to enhance some memorable, CG-enhanced stunts that set the tone for the film. Because some of those stunts would be featured in the film's earliest teasers, the pressure was one from square one. (If you watch the teaser and then the trailer, you'll see two different versions of some of the FX shots-in-progress.) F&V asked Del Conte, lead CG supervisor Andrew Roberts, and composite supervisor Eric Buhwiler to talk in detail about the film’s longest, most elaborate action sequence, which takes place in a long, dark tunnel with traffic coming at our heroes from both directions.
“A lot of shots in the film had CG cars in them, including the tunnel sequence,” said Del Conte. “A lot of the cars that you see close-up when Bruce swerves his car and turns around and heads back down to the other end of the tunnel are CG. We actually added a bunch of CG cars at the front of the car line to give the feel of imminent danger that was needed for the sequence. So the majority of the cars that are close to Bruce Willis’s car weren’t in the actual stunt take – and the tunnel that you see back there is all CG.”
|An example of the “3D tunnel” extension to the underpass where the sequence was photographed. Note the added (CG) cars in the final (bottom) shot.|
“We really put a lot of time and detail into displacement on the tarmac, and using the photo reference for the walls and pillars,” says Roberts. When [Die Hard VFX Supervisor] Pat McClung and [Die Hard VFX Producer] Joe Conmy came in and we showed them our renders, they were blown away. They said, Ã¢Â€Â˜This will be great for an extension – but there are two shots that we didn’t get in the tunnel, and we can use this as the hero background.'”
That came in handy when a practical shot didn’t quite deliver the goods. Del Conte explains that a close-up shot of Willis diving out of his car at the last minute before sending it airborne was originally shot at half-speed on a flatbed truck – and it showed. Digital Dimension was asked to wipe out the live background and replace it with its own CG tunnel and road. “In the end, all that’s real in that shot is Bruce and the car he’s jumping out of. Everything surrounding the car is CG,” Del Conte says.
The team also put lots of effort into making its CG cars not only photoreal, but flexible, so that requests from the director or VFX supervisor to tweak one aspect of a scene were easily answered. “Our process of building and texturing the cars works really well – making sure that all the geometry detail is there, that all the rounds are there so the specular highlights really get hit in a reflective environment, and working with car paint and multiple layers so the compositors can really dial in the level of reflection and ambient occlusion for shading in those areas,” says Roberts. “All those little details that we can get quite anal about really show in the end frame. We provide a large number of passes to the compositors so they can really dial and control a lot of those things and turn out iterations very quickly, rather than us constantly having to re-render an entire scene.”
And, because the tunnel was a 3D environment instead of just an extended background, the interplay between the lights overhead and the reflections traveling across windshields and hoods was perfect, finishing out the photoreal quality of the effects.
“Optimally,” says Bruhwiler, “you’d shoot that with no stuntman in the scene, and then shoot the stuntman on green screen. We had to paint him off the hood of this car with no clean plate available. We had to go to the camera car company – only several months ago! – and find one of several police cars that were used in the shoot and take photos of it to help us clean-plate the stuntman off. For a compositor, it can be pretty thrilling to pull something off like that.”
The very next shot in the sequence shows McClane coming off the hood of the car and hitting the ground. The seemingly simple shot presented its own challenge – as the stuntman rolled across the ground, he had to morph into footage of Willis, photographed separately, mimicking the stuntman’s actions. “During the transition, we had to replace the cop car with a still and then animate the rocking-back-and-forth motion that resulted from the stuntman rolling off the car and also animate some of the lighting changes,” says Bruhwiler. “So the background is a clean plate animated to look like the original plate.”
|The three-way car crash. In the finished shot, bottom, note the addition of our heroes and lots of debris.|
“Because it’s all practical, there’s so much in that shot that you get for free – the way they react, the dust that’s kicked up from the cars,” says Roberts. “It’s a really impressive starting point for us to take it to the finish line.”
“The actors were shot on set again, with a similar camera move and some lighting on them, but you can’t really light actors for this kind of situation, with all the lighting being changed quickly by the cars – and with cars smashing on top of them,” explains Bruhwiler. “We had a lot of in-comp work to bring down overlit highlights, timing it so they looked shadowed and getting some of the spill color from the taxi in there. It was a challenge to put them in the scene. But you had to buy that they were there, or the whole shot was ruined.”
“Also, the shot was pretty clean,” says Del Conte. “I don’t think either of those windshields shattered, so little or no glass fell between the cars, let alone over [the actors]. There was a need for a lot of CG glass and debris flying through the air, covering them and around them, to integrate them into the scene.”
“This was a pretty 2D-heavy shot, but after 3D-tracking the scene in order to figure out where the characters should be, we had to do a match-move of both McClane and Matt,” says Roberts. “We had low-detail characters we used for that purpose, and then we just, frame-by-frame, animated their positions. Then we were able to use those animated characters as collision surfaces for the glass and also matte objects for the glass so that some fragments would disappear behind the CG characters and some would pass in front of them. When comped, it really looks like McClane is integrated into all of this glass flying around and bouncing off him and Matt.”
“We really have to give credit to our roto/paint department because this film is so stunt-heavy,” notes Bruhwiler. “They had a lot of cables and wires, and all of those had to be painted out. The cables on the road pulling these two cars were very complex – we even did some projection mapping of textures to help them paint those out.” Making matters worse, those cables were throwing off sparks as they scraped against the ground during photography. Also removed from the shot was the crane arm and camera hanging off of a visible camera truck, which had to be in frame to capture a dramatic reverse-angle shot leading into the stunt.
Initially, Digital Dimension created a test shot showing cracks appearing on the pillar as the truck hit. According to Roberts, the effect didn’t feel violent enough to put across the force of the crash, so the team started researching real-world disasters like earthquakes to get a real-world reference for what such a crash might actually do to a pillar more than 10 feet in diameter.
“We originally had a little displacement of a section that was going to pop out,” says Roberts. “When we showed it to Len, he said, Ã¢Â€Â˜You know, I really want to see rebar. I want to see pieces blowing out.’ So one of our areas of exploration was just how much stuff should come out of the pillar, and the direction it should fly based on the surface normals of the pillar. Our particle system, Cebas ThinkingParticles, was just looking at the curvature of the pillar and, based on the force of the impact from behind, it was sending fragments out perpendicular to their emitting surface. So fragments would fly off to the sides, but there wasn’t anything flying directly toward camera. We ended up adding some custom elements and fragments that would fill that need.”
Roberts and his team started employing a variety of techniques to fill the screen with debris, adding rebar and chunks of concrete as well as heavy dust and concrete layers. But toward the very end of the process, Wiseman asked them to pull back on the fancy stuff to keep a sense of reality about the shot. “As we added more and more to the shot, the pillar started becoming the event, rather than Bruce and Matt getting out of the way,” Roberts explains. “It became, dare I say, a Bruckheimer moment. We had to pull ourselves back from that, and it was very helpful to have Len communicate what he needed from this shot.” Dust and smoke were banished from the frame, and what survived was a huge pillar cracking and throwing debris at the audience as our heroes, again, narrowly escape death.
“It was a little painful to see all those man-hours not end up on the screen,” Roberts admits. “But it was definitely the right decision in context of the sequence. It would have taken you out of the moment, and you would have concentrated on the wrong thing.”
From big-picture narrative elements to visual grace notes, Wiseman’s view of VFX work was exacting. “We even talked about the lens flare from the headlight of the truck before it smashes out,” says Del Conte. “We had a lens flare we were working with, and he’s like, Ã¢Â€Â˜You know what? I’d rather have a lens flare optically hitting the lens this way, and going across like this.’ We understood what he was talking about, and he was direct about what he was looking for on the photographic end of things.”
|The original shot. Note cables holding the helicopter in place, the lack of rotors, and the equipment scattered on the pavement below.|
|The helicopter pilot jumping to safety was shot separately.|
|The final shot.|
“We had a low-poly representation of what was happening that we could use in each shot,” says Roberts. “We did a 3D track of the flying Crown Vic and had a stand-in for the helicopter. We modeled the head and tail rotors and the engine exhaust. They gave us a lot of reference material for this Astar-model helicopter so we were able to match the rotors exactly.
“We animated rotors with the flying car moving through, and we were then able to see at which frame the car would impact with the rotors. We set up a combination of particle sim and rigid dynamics so the rotors would shear off each time they spun around, as the car got closer and closer. There was a lot of work with fragmentation to make that believable.”
Believability was more important than consistency, the team found, when it noticed that the same simulation didn’t have the same impact when seen from other angles. So additional simulations were run, with the car’s velocity altered and tweaked to fall in a way that seemed credible. Some of the falling debris fragments were adjusted individually, for the same reasons. “Having all the fragments visible against the blue sky, some pieces were distracting – they were falling slowly, and you didn’t get a sense of where they were or how quickly they were moving,” Roberts says. “Some pieces felt like they were floating. We had to bake that simulation and then take pieces out, designing it on a per-shot basis to make the sequence flow.”
Fine, but isn’t there any shot in the sequence that’s all CG? Turns out there is such a thing. Fairly late in the edit, Wiseman realized that he needed a P.O.V. shot to help set up the big helicopter stunt. The shot in question has McClane, frustrated and shaken after the huge pillar explosion, looking toward the end of the tunnel and seeing a helicopter hovering near one of the toll booths. It’s the moment where he decides to send the car airborne to take out the chopper. Unfortunately, that POV shot wasn’t captured on set, and the options available for cheating it were underwhelming.
(Story continues following VFX progression, below.)
|Digital Dimension put together the shot at bottom based on the need for a different POV from the shot available (seen at top). The required perspective was sketched out, and then the existing 3D-tunnel scene was used to build the shot from scratch, using a few live-action elements.|
As elaborate as the final result is, the tunnel sequence only totaled about 25 or 30 of the nearly 200 shots Digital Dimension touched. Outside the tunnel, Digital Dimension did a lot of cityscape tweaks to make Los Angeles locations feel more like Washington, D.C. The helicopter stunt, for instance, takes place in front of several tall buildings, giving the background a very metropolitan feel. That was scaled back a little bit by adding a more modest building – dubbed “Grandma’s holdout building” – amid the skyscrapers. (“She wouldn’t sell to big corporate, so they left her alone,” jokes Del Conte.) And for shots showing an entrance to the tunnel, various buildings and vehicles were removed and/or replaced – because of an odd camera move, one shot was run backwards, necessitating the removal of a moving police car. Digital Fusion was used to projection-map a clean-plate frame over the car after 3D-tracking the shot. (Tell-tale palm treetops had to be removed, as well.)
Digital Dimension’s work on the film also included painstaking shots using Massive crowd-simulation software to visualize an evacuation of the U.S. Capitol, among other impossible sights, as well as some digitally enhanced traffic-jam footage underscoring a plot point about city services being taken offline by a malevolent computer hacker. And there were more shots involving explosive action at a natural gas plant much later in the film.
In the end, Digital Dimension feels like its work here stands apart from the CG pack – even though (or maybe especially because) so many viewers end up completely confused about which elements were live-action and which were VFX. “CG got so overdone in the late 80s and early 90s, with action movie after action movie trying to one-up each other,” says Bruhwiler. “It’s become a poetic, choreographed CG fantasy world, and this just felt – visceral is a great word. There’s a gut-level reaction to it. Going back and looking at the old movies again, seeing Lethal Weapon and the Die Hard movies, I have a greater appreciation for what they were achieving back then, because they had to shoot it all practically. There was no VFX capability or budget for those things. There’s something very human about it.”