How a ‘simple’ driving scene in David Fincher’s film used some significant VFX.
In David Fincher’s black and white Mank, currently streaming on Netflix, there’s a flashback scene in which Herman J. Mankiewicz (Gary Oldman), who would go on to be the screenwriter of Citizen Kane, is being driven down LA’s Wilshire Blvd by his wife Sara (Tuppence Middleton). The year is 1934.
Since that area of Los Angeles has changed so much since then, the filmmakers decided to bring the scene to life with back / rear projection of a car on a stage and a completely digital street-scene backdrop. The CG environment was crafted by Territory Studio.
In this befores & afters discussion about the work, Territory visual effects supervisor Simon Carr and CG supervisor Ashley Pay break down the creation of the Wilshire Blvd backdrop, from previs to final, including why it needed to be such an extensive build, and how it was projected for in-camera final effects shots.
b&a: What was your brief for this sequence?
Simon Carr: So, for a lot of the rest of the film, they could shoot either in LA or around LA, and they had most of what they needed, but there was nowhere to do a significant run down a street of LA and not be huge amounts to remove and replace.
b&a: What were the first things that you did to start that process of working out what you needed to build?
Ashley Pay: We took an initial trip to LA. We knew before we went roughly the area that David was talking about covering. We managed to find some archival footage that David sent us the link to, and we managed to find some other reference material as well before we left, so we had a good idea of the area that it was supposed to be before meeting David. And once that was confirmed, we took a few different cameras out there, I took a bunch of black and white stock for my cameras so that we could take reference photography of what California light looks like in black and white, specifically the look that he was going for, that red filtered look.
We were trying to find buildings at the right time of day, the right kind of style buildings, because there were a couple of them around, which meant we could take some nice reference photography for that to help us later on. There weren’t really enough that we could do any photogrammetry, though. The buildings that are there aren’t what they used to be; they’ve been boarded, and re-signed, and they’ve been re-worked a lot over the years, so nothing is really traditionally 1930s, as it used to be.
Once we worked out the timing of how long it would take to drive, David wanted a minute-ish–we had to work out how long you can drive 30 miles an hour in 1934 down that road? We worked out how far you could travel in a minute and we built 1.2 kilometres of LA and just very roughly, based on Google Maps satellite images, just laid it out so it was to scale. And then we could roughly block in the blocks and the buildings for our initial pass. From then on, whilst I was in LA working there, we could move buildings around together, and pick and choose which ones we wanted to be in the forefront.
b&a: Did you call that a previs? Was it a previs?
Ashley Pay: Essentially, yes. We ran a camera down the road at the right speed and then the DOP gave us the camera angles that he wanted, so we blocked out those cameras with the right lenses, and then we ran a full length run from each camera and gave that to their editor. And then they made a rough cut of the sequence from that, just with the cubes for where the buildings were, using the proxy cameras with a couple of stand-in people for the angles. And then once we had that, it was a case of just building around the shots once it was edited.
They edited the previs and we were pretty close to it in the end. They gave us a tight lockdown on the edit that we had, so we knew exactly what the angles were and when. And as far as I can tell, it hasn’t changed much at all since then. Obviously, when you film it and you cut it, it changes a little bit, but it’s not that different from the original previs at all.
b&a: I understand you did all this with more traditional VFX tools such as Maya—was there any consideration that this could be a game engine / Unreal Engine scene and possibly something played back using real-time rendering on a LED volume? Was that a consideration at all?
Simon Carr: Well, firstly, you need an Unreal team, and actually now I think there are probably more VFX companies with Unreal teams. They’re certainly starting to build, but at the time we started this, it wasn’t really anything that anybody considered for this scene. The advantage with David Fincher is that he tends to be very classic in his camera use anyway, so he knows what he wants and he uses his camera well, so you tend to have the action in the frame rather than being too flashy. He knew what he wanted, and he knows how he wants to shoot the car sequences, so the advantage there was that he was going to lock his cameras really early.
Also, what we were making was a backdrop in a relatively short scene, you don’t want to be spending huge amounts of money on it. So the advantage of him saying, ‘Well, I can lock the cameras down, and then we can lock the edit down,’ means that we were very clear about what would be seen in each shot, and then we could just build that.
If you’re driving along a section of the road and the other side of the road, you never see it, there’s no real point in us building that to any high standard. And so I think that that was the advantage of the previs and the cut, and again, the fact that Fincher knows what he wants, which is something that is not always the case in modern filmmaking.
And then because Ashley is a Maya artist and we have a Maya /Arnold workflow, it made sense that we previs’d it in Maya. And in a way, this is a similar logic to how Unreal is beginning to be used, in terms of the idea that you use the same assets in prep, and then in the shoot, and then in post, and you can work them through the same workflow. And that is essentially what we were doing, but we were doing it in Maya/Arnold. So it was previs’d in Maya, it was built in Maya, it was textured using Substance Painter, lit in Maya, and so everything basically stayed in the same software package all the way through.
b&a: Ashley, when the team was building assets, cars, buildings, roadways, did you have the advantage of any kinds of Lidar scans, or photogrammetry at all?
Ashley Pay: We didn’t use any at all. We scratch-built everything. The intention was to make everything as lightweight and as efficient as possible, so when you look down the road as a whole, not from the camera car, when you look at the street, it’s very much built like a Hollywood backlot. You’ll see two walls of one building. If you don’t see the other walls, we don’t make them. That’s how we built it. We built it to make sure that it was as quick and dynamic as possible to work with. And a lot of the work went into the textures rather than the geometry. You need to get the texture and the feel of LA rather than specifically the geometry, because the geometry of that style of building was very clean. Usually, they’re white rendered buildings or they’re block works, so they’re quite clean to build, and all the nice details come from the textures.
The intention of the look was to feel desert-like, and arid, and hot, so everything was quite bright and clean. At that point, this area of LA is quite new. Everything’s building up for the golden age. Everything’s getting ready to boom. So it was mostly down to keeping as much efficiency as possible in the assets so that we could be flexible.
Luckily LA is built on a grid system, so we designated a block system on the road. If we wanted to move a certain set of buildings to another part of the road, you could just move the block and it would fit in with the road.
b&a: What did you end up delivering or rendering out in terms of camera views?
Simon Carr: Initially, we set up ten shots to deliver but for the final shoot we delivered as eight shots, as four of the shots were combined into two longer runs with some additional work done to join a couple of missing sections. Shot lengths varied, based on the guide cut we’d been given early in the process but the two longest runs were around 700 frames.
Everything was comp’d in Nuke, and we delivered those to David’s production company, and then they prepped them for the actual shoot. We were given a resolution to work to, but I think, certainly at the time that we were working, they weren’t completely sure how the screen setup was going to work. So they did some things like stretching in order to provide extra lighting, and reflection on cars, for instance.
The car was on a platform, and then they turned the car relative to the screen. I think they had a side screen and a front screen, but then they also had LED lighting, so they had light panels above the car for some reflections. It was all shot in color. And it just looped—essentially they just looped the background, and then they’d just re-cue the actors as they needed them to.
The advantage, again, over greenscreen is that when he’s looking out of the side of the car, you can see this billboard coming—which is the point of the scene—so Mank’s actually got time to think, ‘Okay, there’s the billboard, I’m looking at it,’ rather than just being cued and then roughly having to guess where it was, and then us having to time our background to match whatever he acted.
And then he’s smoking, he’s got glasses on, it’s all through glass, there’s a reflective car, all of that stuff which would have to have been done in post is just in-camera. That’s the huge advantage of doing it that way.
b&a: How heavy was this scene you had to create?
Simon Carr: All the foreground buildings and vehicles were modelled and animated in Maya, textured in Substance, with SpeedTree for vegetation, and some simple Houdini crowd. Distant vistas were filled with DMP which was a combination of traditional painting for distance and CG for the deeper mid-ground to allow for the amount of parallax.
This was all then combined in Nuke, with additional texturing, signage and shop fronts also being created and composited by the comp team, alongside the standard depth-cue and polish.
Ashley Pay: I think there’s about 140-odd cars, parked and driving. And there were a lot of trees as well, because it was all eucalyptus trees at the time before palms took over LA. So there’s a lot of greenery to get into as well as a few people walking around in the background.
I can tell you so much stuff about 1930s LA which I never needed to know, but it’s that detail that I actually love, and trying to build those assets to make it as believable as possible. So if we’ve done it right, no-one bats an eyelid at what is essentially quite a throwaway sequence. That scene, it doesn’t really hold down any story points. It doesn’t really control any narrative apart from the significance of the billboard. But apart from that, why would you go to that much effort? Which makes it work so much better for me. It’s transporting you into a time and a location that doesn’t exist anymore without really drawing that much attention to it. And if you can’t really clock it, then it’s worked.