In Your Dreams
Issue: Volume 33 Issue 8: (Aug-Sep 2010)

In Your Dreams

When director Chris Nolan approached visual effects super­visor Paul Franklin of Double Negative with another dream job, Franklin had no idea Nolan meant that literally. Until he read the script. Much of the film Inception, which Nolan wrote, directed, and produced, takes place inside a dream world. The Warner Bros. Pictures production stars Leonardo DiCaprio (Cobb) as a dream thief and Ellen Page (Ariadne) as an architect who becomes a dreamspace designer.  

For Inception, Nolan brought back a team he had worked with on his award-winning films Batman Begins and The Dark Knight: visual effects supervisor Franklin, special effects supervisor Chris Corbould, and editor Lee Smith, all of whom received Oscar nominations for The Dark Knight; stunt coordinator Tom Strothers, who won a Screen Actors Guild Award for The Dark Knight; and cinematographer Wally Pfister, who received Oscar nominations for both of Nolan’s Batman films.

Although the Batman films are comic-book fantasies, Nolan insisted on grounding anything created with computer graphics in physical reality, from streets and buildings to Batmobiles. So, too, the effects in Inception’s dream worlds. “Chris [Nolan] will do his utmost to shoot for real if he can,” Franklin says. “We had fantastic special effects from Chris Corbould and truly astounding stunt work from Tom Struthers, and all three—visual effects, stunts, and special effects—worked in unison.”  

Double Negative (Dneg) provided all the digital visual effects, with New Deal Studios providing miniatures for an alpine sequence. “By modern standards, we didn’t have a huge shot count,” Franklin says. “We had a modest 500 shots; but, it was 34 to 40 minutes of screen time.” Although Dneg altered reality in the real world for a few shots by adding passing views of a landscape outside a bullet train, for example, most of the studio’s effects involved altering environments inside the dream world.

Even so, Nolan wanted those environments to look and feel real. “There’s a large amount of sophisticated visual effects work,” Franklin says, “and what distinguishes it is Chris’s strong aesthetic grounding in reality. He says the audience has to believe it’s a filmed image, not synthetic.” So, even if a shot doesn’t call for a plate, the team will still shoot something real for reference.

Paris in the Dream Time

In the first part of the film, we learn that Cobb, accused of murdering his wife, is on the run. He’s an extractor, someone who uses futuristic dream sharing to invade the subconscious of sleeping targets to steal information from their minds. Corporations hire him to do industrial espionage. But one industrial magnate has a different idea: Rather than steal ideas, he wants Cobb to invade the dreams of a rival and plant an idea. Cobb agrees and hires a team that includes the brilliant student of architecture, Ariadne, who will design dreamspaces the team will share with their target. As she learns how dream sharing works, the first series of effects take place.

We see her and Cobb sitting outside at a café in Paris, discussing dream architecture. Cobb asks how they got there.

Ariadne answers that she can’t remember, and Cobb explains that they’re in a dream. He says people never remember how dreams start, and adds that they’re not in the part of Paris she thinks they’re in. She panics and loses control of the dream world she created in her own mind. When she does, the world explodes.

Nolan shot the scene in Paris. “We were on Rue Bouchut in central Paris,” Franklin says, “a marvelous 19th century classic Parisian street that Chris Corbould rigged with compressed air canisters that fired lightweight debris. Even though it looked dangerous, he rigged it so Leo and Ellen sit in the middle of things exploding around them.” To enhance that practical effect, the camera crew filmed the scene with high-speed cameras at 700 frames per second (fps)—a five-second take slowed to a minute on playback.

“That gave a slow-motion, antigravity look to the debris floating in the air,” Franklin says. “It shows the physics of the world inside their dreaming mind breaking down and falling apart. It’s a stylized look, not like a bomb going off.”

For reference, the filmmakers examined the final scene in Michelangelo Antonioni’s 1970 film Zabriski Point in which Daria, a main character, imagines blowing up her boss’s home. Antonioni represented this by exploding a house, filming it with a high-speed camera, and then projecting it in slow motion. “We wanted to create the same complexity,” Franklin says, “all the tiny fragments you get when you blow up something. But, the practical effects team couldn’t destroy things to the extent Chris [Nolan] wanted. He wanted cobble­stones flying in the air, buildings exploding and shattering. So, we did a lot of CG dynamics. We added levels of detail and complexity to the sequence.”

Franklin began by working with a rough cut that editor Lee Smith assembled in QuickTime from footage shot on the street in Paris, using Dneg’s “Clip,” a Linux-based editing tool. “Clip allows me to draw annotations on top of the sequences and animate them over the timeline,” he says. “I could scribble with a pen, like in Photoshop, to show which bits break up and where the debris would go.”

After sketching an animated sequence in Clip, Franklin discussed his plans with Nolan and Smith via Cinesync’s remote viewing and approval software. Then, keeping notes from Nolan in mind, he replaced the animated drawings with placeholder visual effects animations and sent the new versions back to Nolan and Smith.

A combination of practical explosions enhanced with digital models destroyed this procedurally
created vision of a shattered dream that takes place in a Parisian café.

“The final sequence evolved over six months,” Franklin says. “We’d iterate and iterate and iterate again. We’d review the sequence with Chris regularly and, toward the end, every day. It was an interactive, two-way process, and it was great.”

Effects supervisor Nicola Hoyle led the group that built the models and broke them apart, working with the studio’s DNDynamite, a rigid-body dynamics solver built inside Autodesk’s Maya. Lead effects TD May Leung led the animation team. Modelers used reference photos of material filmed on location to match pieces of debris broken on set. But, they also built polygonal models and broke them apart using the studio’s DNShatter—bits of buildings, cardboard boxes, cobblestones, furniture and tableware from the café, and so forth. DNShatter uses procedurally created patterns based on observed shatter patterns.

“We couldn’t have achieved the level of detail we have in the shot without these procedural tools and a fantastic level of photo­realistic rendering,” Franklin says. “But, we also did a lot of work in our version of Apple’s Shake to re-time the slow-motion footage. The explosions start at 24 fps and then slow down, as if damped by a treacle-like medium, to 1000 fps, and hang in the air. Leo (Cobb) and Ellen (Ariadne) are moving around within a maelstrom of debris flying and shattering around them. It was a great sequence. A brilliant combination of special effects, visual effects, and fantastic compositing.”

Stepping Up

When Cobb and Ariadne return to Paris in her dreams, the burgeoning dreamspace architect more confidently plays with “reality.” In this sequence, which appeared in the trailers, we see the buildings in Paris fold up and arc overhead to create a cube of streets at 90-degree angles, with people walking on the “ceiling.”

The visual effects team combined their digital work with a practical effects technique similar to one that helped Fred Astaire dance up the walls and across the ceiling of his apartment in the 1951 musical Royal Wedding. Franklin previs’d the shot.

As Cobb and Ariadne step up, in effect, onto a vertical plane, they are actually walking on a tilting set built by Corbuold’s crew that has a camera fixed to it. “The whole set pivoted over and they stepped onto the ‘wall’ at the same time,” Franklin says. “We replaced everything except Leo [DiCaprio] and Ellen [Page]. But, we had to get them really stepping up to that wall.”

Because Franklin and Nolan had worked together on previous films, they could reference visual memories shared over the past six years. And for the tilting building shot, they both remembered the drawbridges over the Chicago River. “Chris [Nolan] didn’t want plastic buildings that moved like taffy,” Franklin says. “He wanted a visceral, engineered feeling. In Chicago, the drawbridges lift great sections of road. It looks like the whole end of the street levers up on a giant hinge, so we used that idea. The streets would hinge up and arc over, but they would pivot, not bend.”

To build a Paris street as a full-CG environment, the team from Dneg worked with a visual effects team from Lidar Services that scanned the four-block area around Place Georges Mulot over a period of three weeks. “They had an SUV with a mast that had a

Lidar head on it,” Franklin says. “They digitized the streets down to a quarter-centimeter level of detail and provided us with a high-

detail model. Then the Dneg team did an extensive photo survey of all the buildings.” In addition, because the crew couldn’t get permission to helicopter low over the city, to capture a view looking down onto Paris rooftops, they turned to the Internet for images.

The Internet images, the combination of data from the digital scans and the photographic textures made it possible for CG supervisors Dan Neal and Philippe LePrince and their teams to create the highly detailed digital models of all the buildings. And more. In the shot, everything not on the horizontal, except the actors, is CG, including the buildings, cars, bicyclists, and pedestrians. CG crowds jostle on the sidewalks. “We photographed all the extras, made little 3D models of them, and drove them with a motion-capture library,” Franklin says. “We built a straightforward animation tool set system to place the people and traffic in the shots.”

Paris Light

Although sending people walking up vertical streets and tilting buildings might seem to be the trickiest part of the shot, the challenge was in making everything look photoreal. “If you fold a building over on itself, you can’t see the sun anymore,” Franklin says. “So we had to work out a way to light the street to look natural, yet still fit with the live-action photography. The actors are in broad daylight, lit by the sun. But, we couldn’t have light shining through a building.”

Visual effects artists scanned, modeled, and texture-mapped a four-block area of Paris so they could pivot digital buildings up and around the live actors. A texture-mapping technique developed at Disney helped the Double Negative artists create the photorealistic digital set.

The answer was mystery light sources added artfully. “The lighting had to be so seamless that you never question it,” Franklin says. “It’s a dream that has to feel real. This sequence exemplified the key challenge in all the work: No matter how outlandish the imagery—folding streets or a café blowing up—we had to ground it with convincing, absolute reality.”

The lighting was so complex, in fact, that the crew could not use Spangle, the interactive, in-house lighting tool developed at Dneg for The Dark Knight. Instead, they relied on optimized shaders and raw renderfarm power.

To render the digital buildings and people, Dneg used Pixar’s RenderMan and an updated shader set. “Philippe [LePrince] developed a new shading and texture-mapping system using the Ptex technique invented at Disney,” Franklin says.

Ptex, per-face texture mapping for production rendering, developed by Brent Burley and Dylan Lacewell at Walt Disney Animation Studios, stores a separate texture per quad face of a subdivision control mesh and a per-face adjacency map in one texture file for each surface. First used for Bolt (see “Back to the Future,” November 2008) and “Glago’s Guest” (see “Short Subjects, Big Ideas­­–Simple Truths,” February 2009), the technique uses adjacency data to do seamless anisotropic filtering of multi-resolution textures across surfaces, even those with arbitrary topology. The technique works through RenderMan.

“Basically, with this new way to map textures onto 3D geometry, we didn’t have to go through the process of setting up UV coordinates,” Franklin says. “We used this more sophisticated projection system to map all the textures in the Paris street scenes. We’ve done photorealistic environments before at Double Negative: Gotham City, Chicago. But this sequence happens in broad daylight, in high-contrast sunlight. We had fantastic, beautiful plates shot by Wally Pfister with 64mm clear, anamorphic cameras. There was nowhere to hide. And Chris [Nolan] insisted the digital buildings be equal to photography.”

Compositing supervisor Graham Page placed the live-action actors in the digital environment. But, before he did so, the artists realized that the movement of the actors looked too mechanical. “He worked out a brilliant way to separate the actors even though we filmed them in the same pass,” Franklin says. “And then, he changed the timing to make the shot feel more organic. It was a tremendous piece of compositing.” The compositors all worked at 4k resolution using the tools that Dneg’s R&D department created for the IMAX version of The Dark Knight to overcome Shake’s memory limitations (see “Extreme Effects–Dark Inspiration,” August 2006).

“I don’t want to play down our technical achievements, but this sequence is a testament to the artistry of the crew,” Franklin says.

If You See My Reflection

In the final part of Ariadne’s Paris dream, she and Cobb walk out of the cube and toward the banks of the River Seine. As she approaches a road on the lower level, a bridge springs up out of the ground and builds itself toward her. She then walks onto the bridge.

“The idea sounded brilliant in the script,” Franklin says. “Everyone can imagine it. But building it so that it doesn’t look comical in reality was difficult.”

At top, a digital bridge builds itself toward Ariadne. At bottom, artists at Dneg re-timed the action in the live-action plates to slow down the dream sequence, and then enhanced the image with digital rain.

Nolan shot the sequence on Pont de Bir-Hakeim, the same location used by Louis Malle in Frantic and Bernardo Bertolucci for Last Tango in Paris. “We didn’t have the time or flexibility to build a greenscreen on location,” Franklin says. “Chris [Nolan] and Wally [Pfister] shoot like a whirlwind. So, we rotoscoped the actors off the background using our rotoscope tool Noodle.” The roto­scopers traced around Ellen as she walked toward the bridge, and then the artists replaced the bridge, road, cars, pedestrians, and trees with digital replicas.

The modeling team worked from Lidar scans to build lightweight animation models that animator Dorian Knapp used to drive a more highly detailed final model. “The idea was to have it fold down into the road and then unfold itself, like a mechanical piece of origami,” Franklin says. “Dorian developed the rig as he was animating. He came up with a process by which the ridge unpacks itself and springs into place, and gave it a marvelous clanky, rack-and-pinion feel.”

As Ariadne continues exploring her Parisian dream, she steps onto the side of a platform and pulls on the handle of a giant mirrored door. Then, she steps to the other side and swings around another mirror, resulting in two mirrors parallel to each other. We see Ariadne and Cobb in infinite reflections going back forever, little Leos and Ellens between tunnels of wrought-iron arches. When Ariadne places her hand against a mirror, it shatters to reveal a bridge with repeating arches created from the nested mirror reflection.

On set, Corbould built an 8x16-foot mirrored door on a hinge. The mirror had a large metal rig around it, and the entire unit weighed 800 pounds. “Ellen couldn’t pull the door,” Franklin says. “We had burly special effects guys on the other side pushing. And, this door wasn’t as big as Chris [Nolan] wanted the mirror to be in the film. But, it showed how the reflections move.” Of course, the crew appeared in the reflections, as well.

As he had done before, compositor Graham Page removed DiCaprio and Page from the plates, and, as before, the Dneg artists built a digital background. Neal and lead lighting artist James Benson wrangled the raytraced reflections of the digital bridge, river, trees, and people in the distance. “They studied the way a real mirror moves and added the imperfections that grounded it in reality,” Franklin says. “You think of a mirror as smooth, slick, and seamless, but it isn’t.”

To replicate DiCaprio and Page, the compositors lifted reflections from live-action images shot at various angles, and for a key moment when DiCaprio turns around, inserted a digital double.

School’s Out

The shattered mirror concluded Ariadne’s first lesson in dream work. So, having learned to control her own dreams, she begins designing dreamscapes. When people invade dreams, the dreamer’s subconscious tries to chase them out, and it’s up to the invader and the dreamscape creator to devise evasion routes and methods. So, Ariadne creates looped mazes that resemble Escher’s famous drawings.

“We had to work with the art department to build a physical set,” Franklin says. “The basic idea is well understood, but it works only if the camera precisely lines up with the set. So, we created a carefully designed technical previs that showed the exact camera placement.” Postproduction artists tidied up shots of the actors filmed on the physical set by painting out rigs and painting in holes in the atrium surrounding the stairs cut to accommodate scaffolding.

For the heist, Ariadne designs three levels of dreams inside dreams, all the better to hide in. The trick for Cobb and his team, who will enter their victim’s dream and plant an idea, is that time works differently in the dream world.  Dreams are 10 times faster than the real world, and each dream inside a dream is 10 times faster than the first. “The danger is that at the bottom, Limbo, time runs at a massively accelerated rate, so if you get trapped there, you’re trapped for centuries,” Franklin says. “You would go insane.”

At left, Double Negative artists created the decomposing city by procedurally stacking building blocks inside the shape of a glacier and then crumbled it with a procedural destruction system. At right, the studio sent a blizzard raging through the dream city using its proprietary fluid dynamics system called Squirt.

The first level of Ariadne’s dreamscape is a rain-washed American city. The second level is an elegant hotel with labyrinthine corridors. Third is an Alpine snowscape with a fortress in the mountains. At the bottom is Limbo.

With the help of a flight attendant, Cobb doses his victim with sleeping powder while aboard a flight, and the team successfully enters the industrial magnate’s dream. In the first layer, they’re in a van, with bad guys in a Mercedes SUV chasing them through a rainstorm.

To film the sequence, Corbould rigged a three-block stretch of Wilshire Boulevard in Los Angeles with giant rain sprayers, and Strothers masterminded the car chase and gun battle. “We had to do a lot of manipulation to slow down the action,” Franklin says. “They shot the van skidding around the corner at 24 fps, but Chris wanted it at 700 fps. So, we needed to do incredible deceleration.”

The re-timing software Dneg typically uses, which relies on optical flow to analyze pixel vectors, couldn’t handle the motion-blurred layers of rain. So, compositing supervisor

Julian Gnass rebuilt the shots from scratch by working with a 3D team that extracted elements and re-animated them. “People watch the sequence and think they shot a slow-motion plate,” Franklin says. “But it wouldn’t have been possible. You can’t strap a delicate high-speed camera onto a speeding vehicle.  Muhittin Bilginer [technical director] created layers and layers of slow-motion CG rain falling down, hitting the street, and splashing.”

A Dneg team again added slow-motion CG rain to the end of the shot as the van drops off a bridge and plunges into the river below. For this shot, the artists created the rain using Maya particles, and compositor Scott Pritchard supervised the layering of digital and practical rain elements into the sequence.

Because all the people inside the van except for the driver are dreaming another dream, and because the physics of one layer affect the next, in the second dream layer (which takes place in a hotel), the walls and ceiling move as the van leans around the corner and then freefalls off the bridge.

Borrowing an idea from techniques used to create zero gravity inside a spaceship for Stanley Kubrick’s 2001, Corbould’s team built giant rotating sets. “They built an 80-foot-long section of corridor that could rotate at eight feet per minute to create a tilting bar and hotel room,” Franklin says. “It was staggering to watch and a testament to the immense power of doing effects in camera.” As for postproduction work, the artists’ main task was painting out wires, replacing backgrounds, and adding floating debris. In addition, for shots that take place later on this dream level, the Dneg team put CG faces onto stunt performers.

Below the hotel level, the dreamscape moves outside, to the top of a Canadian mountain, where the art department built a set at 8000 feet and filming took place in freezing temperatures. CG artist Vanessa Boyce led a team that built a CG model of the set and then extended it, adding storms and blizzards using 3D particle and fluid dynamics. Compositing lead Richard Reed’s team fit the model and the effects into plate photography.

And then, with the help of New Deal Studios, they blew it all up. New Deal built a 45-foot-tall miniature and, working from Knapp’s previs, exploded their version of the set and the mountain. “They replicated the action from the previs and took it further,” Franklin says. “And then we added CG bits and more buildings in the background. It was the best of both worlds.”

Inside this dream, the characters snooze their way into a deeper dream, the final level, Limbo.

How Low Can You Go?

Limbo is where dreamers end up if the dream traps them, and at this point in the film, Cobb and Ariadne wash up on the shores of Limbo. Cobb was there before, trapped with his wife for 50 years. Both architects by trade, while there they constructed a modern city, but Cobb has been away for hundreds of dream years, and this city, constructed deep in his mind, is falling apart.

“Chris [Nolan] wanted a city collapsing into the sea, and he wanted it to be completely unique,” Franklin relays. Dneg art director Gurel Mehmet created concept drawings of a sea washing over a city, of a city embedded in a glacier, a sea in city streets, and more, but the drawings didn’t match what Nolan envisioned. “The art department tried, as well,” Franklin adds, “but nothing hit the mark. So Chris concluded that we couldn’t get there with concept art. The idea had to evolve in a complex fashion.”

Franklin remembered that when he was in art school, he would create steel sculptures by cutting pieces of metal that he’d stick together with spot welds. “Using this process, I’d arrive at an end result, a sculpture built from short sections of welded steel that had aspects of a drawing. But this process is the complete antithesis of digital visual effects.”

Franklin sat with Boyle and lighting supervisor Bruno Baron, and they developed a technique that worked. Using reference photos of glaciers, Boyle built polygonal models to capture the basic shape, and then created space-filling algorithms that used basic building blocks. “It was like building a glacier with giant Legos,” Franklin says. “We added a rule to the procedural system to insert streets and intersections, and another set of rules to vary the width of streets and buildings based on the shape of the glacier. Through this iterative process, we ended up with a complex cityscape that had recognizable shapes taken from architectural history, but had a crumbling, decaying feel because it was inspired by the shape of the glacier.” Then, with the help of a procedural destruction system implemented within Side Effects’ Houdini, they collapsed the buildings and destroyed the city.

For a final encounter between Cobb and the ghost of his wife, Dneg created a giant storm that sweeps across the city using the studio’s Squirt fluid dynamics system. The blizzard that tears through the streets and rips buildings apart echoes shots of the collapsing café in Paris at the beginning of the film.

“We are creating outlandish imagery from deep inside the mind with the clarity of a lucid dream,” Franklin says. “Visual effects are an integral part of that. But the most significant thing in this film is not that we are pushing the boundaries of new science, it’s the development of the art of visual effects. We’re reaching a level of sophistication in which filmmakers can treat visual effects as another camera. They can say, ‘I’d like to shoot this,’ and we can film it for them through visual effects.”

Film anything, in fact, that a filmmaker can dream of.

Barbara Robertson is an award-winning writer and a contributing editor for Computer Graphics World. She can be reached at