Divergent: making the mirror room

Perhaps one of the most memorable scenes in Neil Burger’s Divergent is the mirror room, where ‘Divergent’ Beatrice ‘Tris’ Prior (Shailene Woodley) explores a room of thousands of her own reflections. But, it was also one of the most challenging for the pic’s visual effects crew, since acquiring the necessary reflection imagery would have been impossible to film with a real camera and mirrors. Here’s how overall visual effects supervisor Jim Berney, working with visual effects producer Greg Baxter, The Third Floor and visual effects supervisor Matt Dessero and his team at Method Studios, realized the unique shots.

Watch a breakdown of Method Studio’s work for the film.

The problem to solve

“It was a scene that we couldn’t possibly shoot because you’d have infinity cameramen and lights to paint out,” comments Berney. “As well as that, at some point Tris actually comes up to herself and walks around, she breaks the mirror plane and so does the camera. The camera orbits around with her and stops on a two shot of her and herself reflected to infinity. She keeps walking around until she eventually encounters the back of herself as a ‘phantom’ and she turns around and speaks to herself.”

Berney recalls discussing the sequence with director Neil Burger, who asked him whether it was possible. “I said, ‘Of course we can do that. And then I walked away with no idea how to do it at all! But I had the art department make me a small mirrored room with a removable roof and a little mini Tris doll. I sat there and stared at it for weeks, while I tried to prep the rest of the movie. I’d stare at it for an hour a day trying to figure out the story he was telling.”

Watch a part of the mirror room sequence and other shots in the film in this TV spot.

To help plan out the shots, Berney looked to The Third Floor to previs how the scene would play out. “I needed to know how we were going to shoot and execute this,” says Berney. Could we actually do mirrors? Very quickly in Maya we would put in a mirror and another one. And you’re like, ‘holy crap, I see everything!’ And it would be a nightmare to deal with. So not a single mirror was used. It was all shot on greenscreen.”

The other problem to solve was what would be seen in the reflections. Would they be real passes of Woodley, CG passes or some other solution? “Ultimately there are no multi-passes,” explains Berney. “A given take had just have one pass of Shailene shot with multiple cameras. If she crosses herself it’s not like we would shoot her and then get her to move over and do the same thing, because I needed every eye blink, muscle twitch to do the same thing frame by frame and I wanted her eyes to line up perfectly.”

The plan

Once on board, Method Studios refined The Third Floor’s previs to a techvis level. “We went in, adjusted Tris’ height and made it spot on,” says Dessero. “We also adjusted all the camera film backs to make sure they were Alexa 4:3 as that’s what we ended up shooting the sequence with.”

In fact, a setup was devised to use around six Alexas for each shot that would capture a main view of Woodley (plus the dog that later attacks) as well as the necessary reflection angles. The techvis had to account also for the change in room size as the extent of the mirrors becomes clear – from four walls, to six to eight. “Each Alexa camera was basically capturing the reflection of Tris on a particular mirror,” adds Dessero. “And then in compositing, that would get propagated into the next five depths of the reflection.”

A final shot from the sequence.
A final shot from the sequence. Click on the image to enlarge.

The techvis was crucial in determining that around five live action Tris reflections – the closest ones – would be part of that compositing effort, with the further back reflections achieved with CG versions of the actress. “We wanted the resolution, the quality, the performance,” says Dessero. “We never wanted to roto-mate or matchmove all of those expressions, but we could just roto-mate Tris’ performance and do a cloth sim, and that’d buy us all the distant reflections.”

The test

As a result of the previs and techvis, it was determined that there would be about 136 cameras to plot, a daunting task that led to a three day test being devised. “For the test we were only testing four shots, but that’s 24 cameras to hand-plot in a stage that’s a 100 feet by 80 feet,” outlines Dessero. “We did it all manually. We used tape measures and plotted each camera distance in the room. But it became very clear that this was not going to fly on the day. To set up those four cameras it took us over a day – almost a day and a half. We quickly realized that that wasn’t going to be acceptable as we would only have the stage for a short time.”

The solution was to take the techvis, which allowed Method to know where all the cameras had to be in space, and use it to export data to a robotic Leica survey head. “We programmed all those numbers in and went back onto the stage with that data,” says Dessero, “and calibrated the survey head to the greenscreen stage.”


– Director Neil Burger breaks down the mirror room in this New York Times video.

“We plotted out a 10×10 square from the origin,” adds Dessero, “and then we plotted a 20×20 square just to calibrate and make sure everything was working. Once we realized that it was working and we could get the data out of the computer and back into the real world, then we programmed the rest of the cameras into the survey head.”

Method also prepared a shoot plan; essentially a schematics view that overlaid a map of the camera positions and angles on the greenscreen stage with the origin. “We plotted per shot each individual camera on that schematic,” says Dessero. “It had its height and all the pertinent information for the cameraman.”

The shoot

The shoot plan went hand-in-hand with a set of custom 3×3 stickers laid on the ground that described the shot number, game position in x,y and the required height. “We had the greenscreen floor covered in a grid with the camera plots on the ground, where all the camera info on the ground,” says Method’s lead compositor Brian Delmonico, who was also part of the on-set shoot team. “We couldn’t really have that, so we had to mask it, cover them up with green tape until it was time to shoot. It was like a big Advent calendar.”

“Yeah, but with no candy underneath!”, Dessero jokes.

Original plate.
Original plate.

For each shot setup, Method’s on-set crew would drop six cones in the appropriate position, and the camera team – led by DOP Alwin Küchler – would then place the Alexas in accordance with the stickers. The schematics also provided images of what each camera should look like with Tris standing in frame. Woodley’s stand-in assisted with setting up start and end marks. Says Dessero: “We lined up every single camera, approved it, and then we would go through the monitors and if everything looked good, we’d start rolling.”

After each shot was captured, the set would be locked down for around two minutes. “We would survey again each individual camera,” says Dessero, “and we would record all that data so when we got back home we had all this information which was another 3D file which was not only the techvis but also the survey data from the day. We would survey the starting position for Tris, we would survey all of the cameras and if there was a moving camera that boomed from point A to point B, we would survey that also. That gave our tracking department a really good headstart on processing all these scenes.”

Final shot.
Final shot.

The result was that for each shot Method had six greenscreens made up of one hero camera, one hero reflection and four witness reflection cameras, plus photogrammetry elements and HDRIs captured on the set. “This was all labelled and chartered as we had planned in our original techvis and schematics,” notes Dessero. “So we really had a great blueprint on what and how all the pieces of the puzzle came back together.”

Assembling the shots

Armed with the live action plates, Method also embarked on creating the CG components of the mirrored room – the mirrors themselves, floor, ceiling and the distant reflections of Tris. “The approach was to use practical cards for the first five reflections of Tris,” explains Method CG supervisor Blake Sweeney, “but that meant in the V-Ray renders we needed the control over ray-switching, so we could turn off reflections at certain depths. Nothing like that existed at that point so we talked to the Chaos Group and they actually added a few more controls so we could get the full ray depth through mirrors so we could composite the depth pass even through reflections, and it also allowed us to query the ray depth at any given point so we could give the lighters control to turn off Tris to a certain depth before she’d appear.”

“We played with ray depth a lot too,” says Sweeney, “where we started out initially with 16 bounces deep, then we went 32 bounces deep and that wasn’t enough, so everything ended up being rendered at 64 bounces in ray depth.”

Original plate.
Original plate.

Many of the CG elements required specific art direction, including smudges in the glass, occlusion in the corners of the room, dust particulates in the air and the ceiling and floor tiles which were slightly offset and contained imperfections. The lighting of the room, too, was something Method contributed in great detail. “One of the reasons we went with CG into the distance for Tris is that we wanted to art direct the light,” describes Dessero. “The lighting we shot Tris in was very flat because we needed to be able to take these reflection cards and place them and have them not be too distinct or dissimilar. There needed to be a consistent look in the immediate reflections, but after a certain point it becomes a little more impressionistic and we were able to go with pools off in the distance.”

“As the camera turns or Tris moves,” adds Dessero, “you get a ‘swarm of fish’ look and then see that reverberate out in the depth of the scene. That wouldn’t have been possible with flat lighting – so with the pools of light and Tris walking into those in the distant reflections you get a nice vibration of movement into the distance.”

Adding to that look was something Dessero called the ‘banana effect’. “When we set the whole thing up we did it with true mirrors,” he says. “It was perfect cube room, and one of the first creative notes I had on it was that we needed to add some bend to these rooms to get the mirrors to ‘banana’ or bend to the left or right or up or down. So the trick was that we would angle one of the back planes or side planes, depending on the shot, and that would give us the bendy effect in 3D, and we would visualize that and render out a sequence for the compositors to line up to.”

Final shot.
Final shot.

Both Flame and NUKE were utilized by Method in compositing the final scenes. For some of the moving camera shots, artists worked on reconstructing some face and body areas because of occlusions that could not be avoided from the moving camera rigs. “We always roto-mated Tris in part,” says Dessero, “but to get fabric to line up perfectly is not trivial, so we had to sculpt the fabric in a couple of shots and run a sim.”

Compositors also restored an anamorphic look to the final frames to match the rest of the film, since the mirror room had been filmed with spherical lenses. “If we had anything break the side of frame,” explains Dessero, “we would then have to extend the plate or the actor or dog in those scenarios. There was no data for that – it was just reconstruction. Some of the witness cameras gave us that information, and some of the shots we would actually set up two cameras – one that framed it perfectly and one that was further back and slightly offset so we could capture the actress or dog walking into frame.”

Canine effects

Production filmed with a real dog for most of the scene, with Method implementing a CG version for a shot of Tris tackling the animal and then falling through the floor. The studio also carried out dog reconstruction and jaw and tongue replacements for specific shots.

“We prep’d our CG dog for the previs,” says Dessero, “and we built a good looking dog without a ton of facial control because it was running away from the camera and you could only really see it from the side. But we had to build a fur system, muscle system, jiggle controls, bristling of the fur, translucency and tapering of the fur. That work was done in Houdini and rendered through Mantra. We would roto-mate the real dog for the CG one. Usually he was just replaced the second or third trace back – it was just easier and it held up instead of trying to key him.”

Reflecting on the mirror room

“That was the most tricky, challenging and most rewarding sequence I’ve ever done,” says Jim Berney, who makes particular mention of Method in making the shots possible. “I absolutely love it. Method just freakin’ nailed it. It’s pretty damn good. You could give me another year and I wouldn’t have made it any better. It’s done. And that’s hard to say.”

“It’s a difficult scene because ultimately it’s a boring room – a floor, ceiling, mirrors and an actress,” adds Dessero. “So how do you make something like that look complex and interesting? In the end it’s a very cool sequence because you look at it and think how did they film it because you have all these reflections but you don’t see any reflections of cameras in this scene.”

All images and clips © 2014 Summit Entertainment.

1 thought on “<em>Divergent:</em> making the mirror room”

  1. Pingback: 5 – The Space Within Mirrors – Visual Language Case Studies: Objects of Interest

Comments are closed.