article.php

There’s a scene in the latest James Bond movie, Quantum of Solace, where Daniel Craig and Olga Kurylenko get tossed out of an airplane with a single parachute between them. She has the parachute, so they have to hook up in mid-air and then pull the rip chord and descend into a big hole in the ground. The production company, Eon, wanted a realistic freefall effect, but when you do these kinds of bluescreen effects on stage using a body pan, and then composite actors who are freefall skydiving, you always end up either using stuntpeople to do the real falls, or else the actors look like they’re on a body pan with the wind blowing in their face – it’s never quite there. So what Keven Haug, the visual FX designer, as well as the producers, director Marc Forster, and cinematographer Roberto Schaefer decided to do was take advantage of a place in Bedfordshire, just outside London, called Bodyflight, which is an old ex-military site that contains a vertical wind tunnel.

It was built by the British military way back to study the effects of flat spin on aircraft, but it’s since been privatized, and it’s being used as a facility for people who want to try out freefalling and skydiving. There’s this big vertical tube with a net at the bottom and a net at the top, and a 4000 horsepower five-bladed fan above. When they turn it on and dial it up, it creates a 130-140mph updraft. You jump out into it, it lifts you up, and you’re freefalling. So they chose it to help produce a very realistic skydiving effect, with the correct posture for the actors and, more importantly, the ability to use the real actors to do a no-parachute skydiving stunt. It also gave them the ability to do some nice close-ups of the actors.

Kevin Haug came up with a plan whereby they would use multiple cameras in a motion picture equivalent of the bullet-time effect, where they could morph between camera views of the actors skydiving and choose any direction or any POV – any way to look at the actors – so they could lengthen the skydiving effect. The plan was to get as many high resolution cameras in absolute shutter sync into this environment as they could. I’ve done similar kinds of VFX shots with Kevin before, so he asked me to come in and synchronize the multiple cameras.

I had to sync eight Dalsa Origin 4K cameras, seven Sony F900Rs and a handheld, battery-operated wireless Arri 435 film camera, which was being handheld by George Richmond, the show’s operator. He would fly along with the camera in front of Daniel or Olga, getting close-ups of them on 35mm while I was shooting the action with the other synchronized cameras, which would give us a combination of photography and data points in a 3D environment that Double Negative could use to meld real photography of the actors with CGI generated from multiple data points of the actors. All of this was happening at 29.97fps, rather than 24, because they wanted a slight overcrank. Roberto had originally tested some people inside the wind tunnel with a film camera, and he and Kevin told me they were looking for a frame rate of between 28 and 32fps; so I said there’s a perfect window at 29.97 to use the F900Rs. They were the limiting factor in terms of going variable speed. So 29.97 was chosen because that’s what the F900Rs could handle. With the Dalsas I was using their Leica-made anamorphic lenses, designed by Dan Sasaki and custom-barreled by Eric Peterson of A&S Precision. For the F900Rs I used a number of Canon zoom lenses; all matched lenses.

A big part of the challenge was to give the CGI people from Double Negative in London enough depth of field, but also a partially closed shutter; and we agreed that we needed to close the shutter down to at least 90deg. So for me, a big part of the exercise became how to close the mechanical shutters of eight Origins, the electronic shutters of seven F900Rs and the mechanical shutter of an Arriflex 435 film camera all down to 90deg, and then synchronize them for center shutter open sync.

If you want to use these cameras as data points for 3D CG, they have to be synchronized. If, for example, two cameras were 180deg out of sync, the cameras wouldn’t be taking a photograph of the same moment in time, so what the actors were doing would be different in the frame of one camera from the frame of another, which means it would be impossible to derive any meaningful data from the two cameras if they are not synced. If an arm lifts up in a particular direction during a given frame on one camera, and it’s started back down on another camera, then the data can’t be related from one camera to another. It’s like the classic problem of a Picasso painting: I always mused that Picasso never instructed his subjects to sit still, so that every time he looked down to paint, by the time he looked up the subject had moved, and you end up with Picasso’s recognized style. That’s a bit of an exaggeration, of course, but it gives you an idea of how asynchronous photography would look. You can’t triangulate on any given action and recreate it as a 3D CG exercise unless the cameras are synced. But once they are synchronized, you can recreate all the positions fairly readily. It just becomes math. So the viewer won’t know where we’ve mixed between using the photographic images and CGI recreations. It’s pretty groundbreaking.

The Origins have a mechanical shutter with a software function that allows you to issue a phase delay to the center shutter open or the center shutter closed positions in degrees. The 435 has the same functionality in its camera motor, with the exception that I had to deliver sync to that motor wirelessly, so I used clocket boxes that were locked to an Evertz 5600 master world clock, which was locked to GPS. The hardest cameras to change the sync on were the F900Rs, so I determined how to relate the sync-in signal to the F900R to the relative position of its image accumulation cycle. That is, what does a pulse mean to sync on the F900R? I knew that gen lock happens at a specific point in the image cycle and I figured out how to relate that to the effective taking interval of the camera. Once I had determined that, I phased everything to the F900Rs. And how I determined whether we were in phase or not was by taking an F900R and pointing it down the open lens barrel of both of the other two mechanical cameras and just took a picture of the imager. Using the photography from the F900R, I phased all of the Origins and the 435 until they were taking their pictures at the exact same instant as the F900Rs.

The shoot

The first thing I did when I got there was to take a tour of the wind tunnel, which is six stories high. Roberto Schaefer had already scouted the location and had instructed the electrical crew in placing most of the lights, but because the CGI crew needed us to shutter down to 90deg and shoot at a fairly deep stop, I determined that we would need additional light. Using the tour, I scouted every place we could conceivably put lights to get them into the environment. It looked like there were a number of places that would work, much to my surprise, including right alongside the walls of the tunnel itself. It turns out this was a laminar airflow area, and if you pull the viewing ports off the walls of the tunnel, it doesn’t disrupt the flow of air. There was also an area below the unit, where the air changes direction and recirculates upward, and I was able to sneak lights in there, as well as next to the cameras, so I managed to build up a pretty decent stop. But it was a major undertaking. Roberto was a joy to work with under these restrictive circumstances.

We were only at Bodyflight for a couple of days, but I spent a couple of weeks prepping for it, ironing out the sync issues and testing and prepping the cameras. We shipped them to London, reassembled the entire rig and tested it there, then we disassembled the rig, took it to Bedfordshire, and reassembled it again. I think the only hiccups we had were one bad sync cable and a moment when the fan suffered a broken seal. We had to stop for a little while to change the seal, because it was dripping oil.

We had mapped out a predetermined array of slightly larger than a semi-circle in which to align the cameras so they would get a continuous, overlapping and meaningful picture of the action inside this vertical cylinder. That was determined between Kevin, the boffins from Double Negative and me and my camera crew. Using the F900Rs was a good way of getting additional data points using equipment that was sourced locally. The cameras were arranged so that they went: Origin, F900R, Origin, F900R, etc. So they gave us additional data, with the added bonus of giving us some extra photography if we needed it, but it was really just to get more cameras into the same space.

I didn’t bring any of the Dalsas over from France, but I brought over all of those that they had available in the LA office. I recorded them to eight Codex Digital field recorders in uncompressed 16-bit. Somebody, somewhere on the set calculated that we were generating more data per second than Cern particle physics laboratory. But stated in technical units, we generated a boatload. The numbers are off the chart.

The F9000Rs recorded to HDCAM SR decks in 10-bit 4:2:2, which is the biggest signal you can pull of the F900R. I used the SRW recorders in HQ mode, to get the highest possible data rate on those as well. I applied one LUT to all the Codex boxes, and that was dialed in to all the monitors; it was one environment, so there wasn’t much look management to do. I just applied a fairly standard common look for the 10-bit 4:2:2 material coming out of the monitoring outputs of the F900Rs, knowing that it was not so important to see the final look of the picture as it was to see what was on the waveform vectorscopes or the exposure tools for the Dalsa cameras.

We started off the exercise talking about shooting it blue or green screen, but we ended up painting the inside of the wind tunnel white. This was because we were in such a tight semi-circle that the spill problems from blue or green would have been absolutely ridiculous, and at the same time it would have been much more difficult to light. So we went with white and lit it so we could pull a luminance key. Because Daniel and Olga were wearing black, Kevin determined that we would be best to do this with white walls. I spoke to Kevin recently, and he told me they were extremely happy with the shots.

The Origin is an awesome camera. It’s a very quiet sensor. I love working in uncompressed 4K; yes, it is Bayer pattern 4K, but it’s still much greater resolution than HD. It just gives you a lot of bang for your data buck. It’s the biggest signal you can get off a digital camera right now, short of going with the experimental NHK 8K camera, which is a science project – it’s not quite ready for prime time! For me, going uncompressed was important, because it meant I could get all 16 bits off the sensor, which is the lowest noise situation you can get.

It was a great project to work on. Kevin Haug and Leslie McMinn, his VFX producer, did a great job of pulling all this together. The Brits – Double Negative, and the British crew from Bond – were all great to work with. The Bond crew were the most cheerful, helpful crew imaginable. They were just wonderful to me. I showed up with no expectation of what was going to happen, and was immensely pleased with the treatment I received from them. They were so sweet it was unbelievable. Of course it was great having my right-hand man Joe di Gennaro with me, acting as my chief DIT/data acquisition manager. He and I have a sort of shorthand that’s a nod and a wink, and that’s how we get things done. Whenever I try anything like this, I have to take Joe along – I wouldn’t dream of doing anything this difficult without having him there.

Finally, it is a true honor to have worked on a Bond movie, especially with Daniel Craig as 007. I think this represents a new era for the Bond franchise, one that will propel it to success for many years to come!

Blowin' in the wind

David Stump ASC