Double Negative created all of the visual effects for ‘Inception’, roughly 500 shots in the final film. According to VFX Supervisor Paul Franklin, nearly all of them required tight collaboration between the special effects and visual effects teams.From Digital Media World Magazine


Paul had already worked with Director Chris Nolan on two of the Batman films, ‘Batman Begins’ and ‘The Dark Knight’. “Chris always demands absolute photorealism in effects. In this case, the characters as dreamers manipulate the world around them in order to carry out the ‘dream heist’ they are committed to setting up. Making unreal events feel as real as our dreams feel was very challenging.”

As unreal as the effects sequences appear, he noted that what they all have in common is Chris’ preference for capturing as many real elements as possible in camera, such as massive special effects and stunt performances, before deciding on what visual effects will be needed to extend and enhance a scene. The team then adopted this style as a cue for any shots that had to be completely CG.

To produce a protracted sequence in which a van holding all of the film’s main characters tumbles slowly from a bridge into a river in super slow motion, the special effects supervisor Chris Corbould and his team rigged real vans and fired them through the crash barrier off the Commodore Heim Bridge at Long Beach near LA, an old bridge built in the 1940s that Paul described as looking like “the Eiffel Tower turned on its side and painted green”.

These falling vehicles were shot with a Photo-Sonic 4ER, a film camera that loads standard film but can shoot at frame rates up to 1200fps, and a Phantom digital camera, able to shoot up to 4000fps. “They typically shot at 360 to 700fps, which meant time could be slowed down to a crawl, giving the impression of a van suspended without gravity. This was important to show the variable rates of time from one part of the dream to another.”

Slower Than Slow

When Chris wanted to slow the fall even further than the film rate would allow, the visual effects team retimed the footage inside Shake to give a result that could have been achieved with footage shot at 2000fps. The scene was meant to take place in rain, but since practical rain would have been impossible to set up over the set, the team composited CG rain over the top of the scene as a particle animation, contributing to the impression of slowed time. The particles were generated in Houdini and Maya and with RenderMan and Double Negative’s renderer, DNB.

“In preproduction, we had experimented with how far we could slow down footage shot at different frame rates,” Paul said. “Tests shot with normal and high-speed cameras were taken into Shake. We found that shots of clearly defined shapes against clear backgrounds worked well. The software locked onto the shapes and let us slow the frame rate right down. But when the original shot contained motion blur, the results retained the blur, which usually had to be painted out.”

Crossing Layers
What caused Shake the most problems were shots with many crossing layers, such as when Cobb, the main character, splashes into a tub of water. Likewise, a shot showing a car skidding around a corner in a rainstorm, throwing the passengers around inside, proved to be a nightmare for the team. It had been captured at 24fps from a hard mount on the nose of a following vehicle, included practical rain, lots of motion blur and bouncing cars. Chris asked them to slow it down to 700fps.

Paul said, “We had three shots like this. When we tried to retime the footage the layers simply fell apart, and we wound up having to mostly re-build them in CG by projecting the vehicle onto geometry and adding layers and layers of CG rain and splashes. So, we were often figuring out how to proceed as we went along, and had to break some rules.”
Chris tended to avoid previs, feeling that it prevents spontaneity and experimentation. His DP Wally Pfister felt the same way and likes improvisation. In the Batman films, previs was mostly used as a conceptual design tool to set out the story for each sequence, but not for laying out shots. For ‘Inception’, however, the artists knew they would be working on a lot of precise, technical setups, especially for the forced perspective shots. Chris began to understand the value of previs and how much the team relied on it.

Penrose Steps
For example, toward the beginning of the film was a sequence they called the ‘Impossible Steps’. The character Arthur wants to teach the student Ariadne how to create loops inside the architecture, creating an optical trick based on the Penrose Steps illusion. “We built one of these loops as a full-size staircase that the actors could walk on – although it only appeared to loop back on itself from one angle of course,” said Paul. “We wanted to build it as large as possible, but fitting into the atrium of an existing building near London.”

In previs, Double Negative worked out exactly where to place the camera and the exact dimensions of the set. They gave these to the Art Department, who built the set to precisely match their measurements. They also liaised with the camera department and refined the camera placement and the type of crane to use to capture the set at the correct angle. Accuracy was critical. They decided on a 50ft Super Techno crane which, when fully extended, left only about an inch clearance under the ceiling of the building.

“Without having done a full laser survey and comprehensive previs, I really don’t think we could have achieved these shots. We knew just how it would look by simulating the lenses as well. Thus, previs became useful for such shots. We also used a similar approach for VFX shot layout – what now is called post-vis – that allowed Chris to approach shots differently after the shoot. He might say, ‘We need a shot that does THIS, but our footage looks like THAT. It hasn’t been shot for it all. What can we put together?’”

Cubic Paris
Using previs, they could quickly formulate ideas. For Paul, the best example is perhaps the sequence of folding streets in Paris, which grew from a single line in the script, “Ariadne folds the street into a giant cube, and they start walking around on the inside . . .”
The Art Dept had produced a few concepts showing how a closed cube composed of Parisian streets might look, but nothing to show how they might have arrived at that condition – how did the streets fold up, or bend themselves around the scene? Chris was open to ideas but one look he especially wanted to avoid was bending the buildings, as though they were made of rubber. He wanted the movement to have weight, as though a giant mechanism were driving the action.

“Before shooting ‘Batman Begins’, we had spent time on the streets of Chicago, looking at drawbridges across the Chicago River. When these bridges fold up, they lift the entire road surface with the sidewalk, street lamps and furniture all plugged into the bridge deck. If you are standing at the opposite end of the street, the whole road appears to be folding over. ‘What if,’ we asked ourselves, ‘it continued folding right overhead, hinging a second time into a giant cube?’” Paul said.

Nowhere To Hide
“To begin, we made a simple model of the Paris location, based on dimensions derived from Google Maps satellite photos and reference images. One of our artists made a quick previs of it arcing overhead, and Chris immediately said, ‘That’s it!’ No more previs was required, but he did take it on location as a basic guide for the actors as well, telling them where they should look, for example, as the scene proceeded.

They had to make sure that, once they had the sequence in post, they would have every detail they would need to build the folded streets to match Wally Pfister’s sharp, clear, high-contrast photography. “Everything happened in bright daylight as well, unlike Batman – we had nowhere to hide in terms of detail, no shadows even. We had to reach a level photo-realism we’d barely touched before.”

Before the shoot, the Double Negative crew spent two or three weeks on location with Lidar VFX Services, a company from Seattle, to document to area, photographing and taking survey measurements and Lidar scans of about four blocks of the city. “We simply didn’t know at that stage exactly what Chris would want to do. We wanted enough documentation for a total reconstruction for unexpected bridging shots, for example, that may not have been recorded on film.

Texture Mapping
“For texture painting, the next step, the shader writers used the Open Source application Ptex for texture mapping, made at Disney Animation. It means you don’t have to assign UV coordinates to every point on the surface of your 3D model in order to attach textures to it. Implementing it takes some skill, but it did speed up the texturing process considerably and let us spend more time on perfecting our photorealism and less on the technicalities.”
Lighting the scene was another hurdle. If you think about the streets closing overhead, shouldn’t the sunlight be blocked out? The team had to work on the assumption that this closed universe would include gaps for light to get through, which the viewer can see if he looks carefully. They also had to allow the position of the sun to change over the course of a couple of shots. After all, they were inside a dream – shouldn’t unexpected things happen? In any case, to make the scene go dark would have required a night shoot.
“Of course, the problem of creating these folding streets wasn’t entirely confined to buildings,” said Paul. “The people you see walking on any but a horizontal surface had to be generated by a crowd system using models based on extras shot on location in Paris. We had to build a whole library of European cars as well because, although we’d built cars for other projects, including the Batman films, they had been American cars.”

Face to Face
The Paris sequence continues on to a scene out on a bridge, which Ariadne has created in her dream from a series of reflections between two mirrors facing each other, bringing the team a new set of challenges. Because it was based on a real bridge shot in Paris, the Bir Hakeim footbridge, again the whole location had to be surveyed. “When we were out scouting the location, I remarked to Chris that it reminded me of a hall of mirrors with its rows and rows of arches repeating on across the river like the flections between two mirrors. He like the idea and wondered if we could literally take a big mirror with us to the location.

“Chris Corbould’s team built a huge mirrored door 8ft by 16 ft that swung on a giant hinge to load into position and produce reflections of the actors to work with. But, of course, that was only one mirror. We could never have managed two – it was that big and heavy to get into place. Also, in the scheme of things, it wasn’t even as big as we really needed. A mirror of about 20ft by 40ft would have been required to cover the scene, but the one we had already weighed 800 lbs and took several men to push into place.”

Reflecting Reality

However, it did give them the reality of a real reflection in a real mirror – the way it wobbled and flexed, and the lines between the panels. It supported Chris’ preference for grounding effects in observed reality, not perfection but something that felt real, taken into a place that felt real. Those little imperfections, captured in the photography, are what sell the effect to the audience, who can believe they really had giant mirrors there.

“It also reflected everything, not just Cobb and Ariadne but also the camera crew and equipment, visible in all of our plates. We ended up having to rotoscope the actors completely off the backgrounds and replace everything in the reflections around them with a CG riverside environment. Then we created all the reflections between the mirrors, as raytraced reflections of a detailed model of the bridge.

“In fact, even Cobb and Ariadne are mostly CG doubles because we didn’t have clean reflections of them in the mirrors. Fortunately, they were mostly out of focus so the point where live action changes to CG doubles is invisible. With each reflection, as well, you lose about a half a stop of exposure, causing them to recede into darkness. We had to cheat it a little to still show a recognizable bridge when they shattered the mirror.

Fragmented Physics
Render times were especially heavy, on average about the heaviest they’ve done at Double Negative, due to the level of detail. Also, Chris wanted all effects completed at a minimum of 4K resolution, and some were done at 6 or 7K. They also made special IMAX versions of about 25 shots.

The explosions in Paris were another special effects/visual effects collaboration. Chris Corbould worked out how to stage ‘fake’ explosions along the street with compressed nitrogen cannons firing debris up into the air, such as fruit made of foam, paper and other lightweight objects. These were shot with high-speed cameras that Paul’s team used to create a floating look, as if hanging in mid-air.

“Ariadne, in her panic, had lost control of the physics in her dream,” explained Paul. “The laws of nature break down and everything fragments. It wasn’t to look as though a bomb had gone off, just a disintegration of physics letting everything fly apart.” The high-speed footage was effective but they couldn’t fill the entire street with practical objects nor sustain them for very long without CG. More important, they couldn’t have involved more substantial objects like bottles, cans and cobblestones from the street, in the same way.

Recursive Explosions

Working with the speed of the explosions was important. “Everything starts in real time but slows right down, as if subjected to enormous drag, or underwater. Chris didn’t simply want it to look like slow motion. So we introduced secondary, recursive explosions of objects already flung into the air, eventually breaking down into dust. It was a great effect to work on. We talked about it a lot and knew the live action was good but would never be enough.
“It evolved over the post period. I worked out a rough cut of the sequence with the footage and drew on it with our software called Clip that lets you make animated notes directly onto the footage – good for working with editors and directors. It helps on a playful sequence like this that everyone would recognize as CG but that had to look and feel realistic.”

Gravity Revolution
A dramatic fight sequence in the ‘Inception’ story takes place inside a hotel that begins to rotate in response to the van, holding all of the characters, as it tumbles down a hillside. As fighting breaks out and the actors run up and down the walls and ceilings of the room and corridor, all action was captured in camera filmed in a special set, about 90ft long, that rotated a full 360° and continued spinning at six revolutions per minute.

“The crew worked out how to attach the camera inside the set, moving back and forth on a dolly, which gives the fighting great dynamism and was inspired partly by Stanley Kubrick’s work on ‘2001’ for which rotating sets were also built,” said Paul. “The difference here was the dynamism created by the movement of the camera and its use in a high action scene.

“However, when the scene changes to full zero-gravity, visual effects were needed to augment the practical effects. For this we built a vertical version of the corridor, that is, on its end, and lowered the actors down into it on wires with the camera down at the bottom pointing up at them.” The actors mostly hid their own wires, but the stunt rigs were very evident as they bounced around and needed some challenging rig removal due to the fast-paced action. Any visible wires needed very precise roto’ing to paint out and restore the background.

Fighting Fit
“We made a CG model of the interior of the corridor and, under zero-gravity, we added floating, bouncing objects to help sell the impression of weightlessness and add to the confusion about which way is up. No digital doubles were required, only head replacements for two stunt actors. In fact, actor Joe Gordon Levitt trained intensively to be able to perform these scenes himself.”

Paul thinks that capturing this scene as live action gives it an immediacy and a kind of ‘messiness’ that would have been missing from a CG recreation, though it may have been just as exciting. It might be compared to the elegant, highly stylized fighting in ‘The Matrix’.

Living in Limbo
Limbo City is the final destination of the film’s dream, where Cobb and his wife Mal have become trapped. Time is accelerate within the dream, just a few hours in the real world represent years inside the dream. Cobb and Mal had been there for 50 years spending the time building an immense modernist-style city together. Both of them love architecture and Cobb is architecturally trained.

The design of the city arose partly from Chris’ and Paul’s own interest in early 20th century modernist architecture. Though the buildings do not exist in reality, they are inspired by such architects as Le Corbusier, Mies van der Rohe and Walter Gropias. “Limbo City, in one respect, represents the history of modern architecture. Starting on the outskirts, you have early 1920s modernism and Bauhaus influenced designs, Russian constructivism and the International Style. But as you walk through it, you pass through all the different stages of architecture until you reach the centre with colossal towering buildings, still inspired by Le Corbusier but taken to an extreme degree, a thousand feet high, for example.

“In the dream, so much time has passed since Cobb and Mal built the city that it’s now crumbling back into the sea and serves as an indicator of Cobb’s subconscious mind and mental state. Chris didn’t want the buildings to simply appear derelict. He wanted them to be monumental, with aspects of natural land forms like glaciers or cliffs.”

Sculptural Approach
The design process arriving at the final look was convoluted and approached like a sculpture. They wrote a piece of software, using Houdini and Double Negative’s fluid system DN Squirt for splashes and dust, that would start with a simple model of a glacier based on a photo reference but filled it with architectural blocks, resulting in a ‘Leggo’ version of the glacier, with the buildings rising and falling as a glacier would do.
Once the shape was right they added rules, such as – ‘every four blocks, add a street; every four streets, add an intersection; vary the widths of buildings according the slope rising above, and add damage according to the ravines cut into the glacier’.”

Refining this software took a few months of iteration to produce the models, and resulted in a city with a complex geometric architectural layout that, seen from a distance, had an organic quality, looking like a series of eroding cliffs. To finish, they added collapsing buildings, not falling the way real buildings do when demolished but the way icebergs break off the face of a glacier, pulled by weight and affected by air resistance. The team spent lots of time looking at demolished buildings, collapsed, damaged, derelict housing projects to give it an eerie, apocalyptic feeling as well.

Limbo Square
Deeper into Limbo City where the damage is less pronounced, confined mostly to broken windows, we encounter Cobb pointing out specific buildings on Limbo Square, encompassing diverse architectural styles and different houses he and Mal had lived in, based on various sources. Mal’s childhood home is based on a real house located near Paris, which was photographed and made into a digital model. Others were modelled from buildings Double Negative sourced themselves.

But how could they be presented without looking like a mismatched collection of buildings in a theme park? They shot plates of a large reflecting pool at the Department of Water and Power in LA and used it to position the building as though they were floating in the water around this pool. They started with the real pool but had to replace most of the water digitally and made it much larger to accommodate all the buildings. The 3D artists spent a huge amount of time detailing, adding damage and dereliction and, above all, photorealistic lighting.

Paul reflected on the project. “The most exciting part for me about working on this film was the chance to work with a director like Chris. He’s very demanding and remains in total control of his films as writer, director, producer – in short, an auteur. But he is also very collaborative, listens to everyone’s ideas and uses any idea that is genuinely good. In consequence, the team members have a creative investment in the film, and do their best work.”

Massive Miniature
Chris Nolan and the production decided that the destruction of a massive concrete fortress set in a snowy, mountainous landscape would be best portrayed with a large miniature. After having worked before with Paul Franklin on ‘League of Extraordinary Gentlemen’ and with both Chris and Paul on ‘The Dark Knight’, Project Supervisor at New Deal Studios Ian Hunter, based in LA, agreed that Chris likes his effects to include as many real, in-camera elements as possible, but is not opposed to CG as a tool. “He takes effects in levels, starting with full size physical models. In the case of the fortress, that would simply have been too massive. But a miniature was the next level, still in camera with practical explosions, just at a smaller scale. This way, photography was still driving the effects.”
Ian’s team made sure that the scale, lighting, textures and paintwork on their miniature matched the full set in Calgary, Canada perfectly. They also took care that their miniature’s explosions had the same texture, colour and feeling as Corbould’s full explosions. Again, live action was guiding and driving the visual effects.

Eye-Popping Shots

Part of New Deal’s work is to photograph their models as well as build them. A demanding shot had to appear to have been captured from a helicopter, which meant figuring out the correct height, how long such a shot could realistically be sustained, the speed of the camera motion and other details that convince the audience that it is a real helicopter shot. “Those eye-popping virtual camera shots are exciting,” Ian said. “But they can pull a viewer out of the story. Virtual cameras are sometimes too good. People intuitively recognise views that they can never experience, preventing that vicarious feeling of ‘being there’. The ‘folding Paris’ shots are sheer fantasy but we don’t reject them because they are shot from the point of view of a person on the ground.”

The set in Calgary represented a portion of the fortress at full scale. Ian went on location to carry out a photographic survey, and take measurements and lighting reference. “It was solidly made of timber, finished to closely resemble concrete and with real icicles. Our miniature represented about 60 percent of the real building, the set showed about five per cent. Any full establishing shots of it were provided as matte paintings from Double Negative.”

Shared Previs
Designs for the structure came from the Art Department, prepared in Rhino 3D modelling software, which New Deal also uses. They compared these with the set’s blueprints to ensure that doors, windows, railing and other small details found on set would match the miniature and appear to scale. People, for example, had to be shown standing near details of the same proportion as those on the miniature. The same Rhino 3D model was used for previs at both New Deal and Double Negative on the shot itself. This meant that, at the start of the build, they could base their construction drawings on the same digital assets as the Double Negative team was using for VFX shots.

An important reason for this consistency across the production is time. As they developed and planned the shot – angles, approaches and so on – with Paul’s team in the animation phase of the collapse, his crew could meanwhile keep building. By the time Double Negative’s final previs was approved, New Deal could be preparing pieces of their model for assembly without concern about deviating from the plans. Working everything out digitally first before committing to materials also keeps them close to the Art Department.

Minimum Scale

The model was built at 1/6 scale and consequently was not especially ‘miniature’. The actual fortress in the story was massive, and depicting destruction and pyrotechnics typically demands a minimum scale to be convincing. Also, certain materials will only break apart effectively when the pieces are large enough. Striking a balance between results, cost and practicality is tricky.

Another compromise came with the venue for the explosion. They would ideally have staged it with more space but the budget didn’t cover relocating the model. So they kept it in the back lot of their premises. Also, the scale and amount of the fortress and mountainside they built was dictated partly by space, and partly by how much they needed to produce the right look and interaction for the collapse.

Everything else in the shot could be extended with digital matte paintings by Double Negative. Paul shot lots of mountain reference, as Ian did, so both teams’ rocks, for example, matched. The model breaks apart onto their model mountain, but surrounding it is green screen to give Double Negative some scope to extend the shot out. But everything involving the break up and fall down the mountain was achieved in miniature. They aimed to maximise the available model to create the best effect.

Shot Planning

In previs they had placed cameras, five in all, in locations they thought would work well, serving as an approximate guide for how extensively they detailed the model in certain parts. For example, the buildings at the back would be seen in detail from mainly two and a half sides while the tower at the front tumbles down the and was seen on all four sides. They can be addressing these details as soon as the previs is approved, due the accurate 3D model.

The y shot everything with Vista Vision film cameras, which have a larger format than a regular 35mm camera and will hold up for enlargements filling the IMAX screen. However, the highest frame rate of these cameras is 72fps. Miniatures are typically shot at a higher rate to be able to slow the footage. This limitation was one more factor in choosing the 1/6 scale for the miniature.

“New Deal takes a lot of care to make their shoot match first unit expectations,” Ian explained. “We need to ensure that the miniature performs as shown in the previs. Therefore, much of the destruction was not a result of the pyrotechnics and explosives but of mechanical effects built into the model. These were tested extensively as well. In this case Chris Nolan had a precise vision of the collapse. The tower at the front should collapse first, followed by the buildings at the back. It was also meant to be a controlled demolition in the story – characters are seen planting bombs along the bottom, causing a fall from the bottom up, and from the front to the back.

“To control this, while the pyro went off producing the fire, explosion and debris, inside we had hydraulic scissor-jacks, like elevators lowering the building, and all floors were attached structurally on trips that could be released in sequence.

The building could thus be collapsed with out explosives. Even better, we could have another go should things not look quite right the first time – which is exactly what we did. Although we did have to make two sets of outer skins for the buildings, the steel internal structures could be re-used.”
The collapse comprised about 200 small events – explosions, cable releases, dropping elevators – all choreographed and timed out with electronic timers to 0.001 second. While the real event, coordinated together took 5 1/2 seconds, on screen it lasted about three times that long. Retiming gave it the correct gravity but also, as the back buildings follow the front, an explosion and lighting was timed to go off behind the building as it fell. The previs was critical to working out, frame by frame, where explosions and events should be placed and how to time them. They needed the two takes but parts of each were useful, due to careful planning.

Timing Boxes
The Calgary set was built in timber coated with a concrete-like surface and was never intended to be blown up. New Deal’s model was a much more fragile construction of plaster over polyurethane foam, with the steel structures inside. The plaster would crack and break like concrete while the lightweight foam supported it.
The timing was controlled with pyro-technicians’ electronic solid state timing boxes with 24 programmable timers each. “We used four of the boxes for the 200 or so events, each triggering the next from box to box. Fine electronic timing is necessary because events like fire, trips and break-ups happen naturally at different speeds, but must be recorded on camera in a precise order.

Perfect Light

Between the two takes, eight different camera angles were sent to editorial. The ‘helicopter shot’ was important, and had to match the actual helicopter shot done in Canada accurately enough to intercut the footage. The bottom of their miniature was in fact 17ft off the ground to allow for the fall down the mountain, and extended 40ft up at the top of the tower. For the helicopter effect they needed to shoot from another 30 ft above this. They used a techno crane with a 50ft arm to give the correct swinging effect but to lift the crane a full 70ft off the ground, they built a platform for it and raised it with a construction crane. Thus they could match the style of the real helicopter shot.

Lighting was a genuine concern. Paul advised them that most of the live action was being shot three-quarter back-lit. To plan the best position for their model, they also used a 3D model of their back lot where the shot would take place, complete with sunlight angles at different times. Based on best information, they positioned the miniature and scheduled the shoot.

However, on the day, the live action shoot in Canada was shot right after a snow storm under suffused light. By great good fortune, the LA afternoon New Deal had planned their own shoot for also turned up overcast. “I was worried, but actually my cool-headed Producer David Sanger assured me it would work out. Otherwise,” Ian admitted, “I’m not sure how we would have proceeded.”

‘Inception’ was certainly among New Deal’s largest, most complex projects to date. Their work on this project took 14 weeks, from the design stage with the Art Department to the second shoot. At the peak of construction, it took 60 people to build. “Others have taken longer and involved more models, but the complexity and physical mass of the ‘Inception’ model stands out, plus the 200 events synchronised precisely into 5 ½ seconds and the pressure to produce footage that could be intercut with the first unit’s helicopter shot.
“I absolutely depended of course on the expertise of the team here - Construction Supervisor Forest Fischer, Mechanical Effects Supervisor Scott Beverly, Special Effects and Logistics expert Robert Spurlock, John Cazin on Pyrotechnics and, planning how to capture it all, Director of Photography Tim Angulo.”

Words: Adriene Hurst
Images courtesy of Warner Bros Pictures
Featured in Digital Media World. Subscribe to the print edition of the magazine and receive the full story with all the images delivered to you.Only$79 per year.
PDF version only $29 per year