Sometimes you never know just how complicated things can get.
Whenever I'm called on to do a visual effects shot for someone, my normal reaction is to say "Sure, I can do that" and about two thirds of the time it's true, I know exactly how to do it and that I'm capable of it with my current level of skills and software. But the other third of the time, I am secretly thinking "I have no idea how to do this!" and will rush out to do some research to figure some things out - can the software do it, do I need additional resources, am I really able to do this, how complicated will it be?
There's a shot in Checkpoint that I've been working on, off and on, in various pieces, over a period of six months. It's a simple shot - we filmed our hero on a bridge over a railway track, and had planned to do a crane shot over the wall of the bridge showing a train on the tracks, with a railway gun. So my original plan was to comp in the train, maybe replace the gumtrees with European pine.
Our attempts to do a crane shot handheld were a bit of a disaster, we really ought to have thought it through a bit better and borrowed a crane to achieve the shot. And we ought to have picked a day with nicer weather, too. But our hands were tied, as is so often the case, and we got what we could, as best we could.
So now it's my job to fix it.
The train model was already long completed, so that wasn't a problem, though it did take me a long time to build.
The first thing we knew we had to do was stabilise the shot as we follow our hero crawling along the wall, because it shakes all over the place. The problem is, with stabilisation comes distortion and motion blur. We can't really do anything about that, though, so I have reduced the shake down gradually as the shot progresses, so that the distortions are minimal.
Next thing we knew was that we would need a 3D representation of the railway line and valley, which had to be a matte painting including distant fields of the French coast, a complete replacement of the trees with European pine, and a French village. So I assembled the elements into a 2D matte image, using the best shot we had of the railway line, but it wasn't good enough for the final shot, and we knew we had to go back to the location to photograph high resolution images of the valley.
The day we arrived was a beautifully sunny day, and the shots of the valley we took were great, but as it turned out, not perfect, because as we crane over and tilt the virtual camera in the 3D environment, it distorted the verticals and all the trees looked like they were leaning over. The answer was to go back again and re-photograph the valley from a slight angle, so that the distortion would be eliminated, achieved by having the final angle of the shot match the actual angle the image was taken, and not fake it like the first try.
One of the issues I had with the crane move through the valley, though, was getting the sense of distance for the French coastal part of the shot. In reality it was a photo of a farm in Holland, but I had it comped in and needed it to move relative to the camera so it felt like it was miles away, and not a flat image. This meant finding ways to have the fiddly crinkly edges of the trees blend in with the background without distorting or obscuring it as it moved. So I had to mask them out using alpha masks, careful to keep the bleeding edge suitably coloured to blend, and move all elements to give a sense of depth. Rendering each layer separately was the only way to achieve this.
The good thing with this approach, though, is that I didn't need to use a tracking program to recreate the environment, and 'lock' the train model onto the real tracks. Instead I am able to apply the train actually to the virtual environment, because they're both 3D models in the same program, and are fixed entities. That is a relief, because 3D tracking is still not perfect, and is a bitch to get right.
The next thing I realised we needed was the wall of the bridge. It isn't enough to use a 2D image and animate it, I needed to have a 3D representation of the wall so that the depth as we craned over it felt realistic. I tried to find a way around this, because I was afraid it wasn't going to work, but in the end it was unavoidable. I built the wall, and the uneven rocky surface it possessed, and applied the textures of the real location that I took pics of. The end result is the wall looks completely convincing, and the crane move is seamless, so it was well worth it.
However, big problem. The original footage was taken on a grey overcast day, and the other photographs we took at later dates were in glorious sunlight. They don't match. I have to apply this footage onto the surface of my wall, while the virtual camera watches it and cranes over it. If the footage and the wall photos don't match exactly, it will look wrong.
I had to alter the colour of the wall image so it matched a grey day, then mask around it with a soft edged blend to overlap with the original footage, so you cannot see the mix between them.
Then I have to render the bridge part of the shot separately to the valley part of the shot, so that when Rob colour grades the two, he can either reduce the vibrancy of the sunny day valley, or increase the vibrancy of the grey day bridge, so that they blend seamlessly. I also have to provide him with an alpha channel mask of the footage so that he can separate the two shots when he grades them.
The train has to be rendered at high resolution. Soldiers have to be added, probably digital ones because we don't have any footage of live ones. Steam coming from the train, smoke from one of the stacks, atmospheric haze, and birds flying, all will add life to the finished shot and give it the energy to make it convincing.
It is the most complicated thing I have ever done. So many elements, so many techniques, so much preparation, and yet every single part of it I was absolutely confident about being capable of achieving it. Individually, anyway. If I'd realised just how complicated I was making it for myself, I would've come up with a different idea for the shot and convinced Rob to do that instead.
Like hire a real train.
4 days ago