A while back, I mentioned my idea for a video piece to accompany Arvo Pärt's Tabula Rasa. After exploring a number of technical and aesthetic options for achieving the effect I had in mind, I finally finished this video at the end of this past week. In this post, I'll talk about some of the compositing experiments I conducted as part of this process, but first, here's the final video:
In the end, the procedure I used was very similar to the basic sketches I showed in my earlier post: embedding a still photograph of the white plastic car in the space of a video. However I learned a number of things about how better to do this by trying a few wildly different approaches first.
My initial composite tests had two major flaws: poor perspective matching and poor lighting matching. The white car never really looked like it was sitting on the street and it didn't look like it was affected by the lights around it. To work around these problems, I first attempted to achieve my shots in-camera by using traditional forced perspective techniques. Before heading out into the night street, I tested this technique in a controlled environment:
This first example is the closest to a true in-camera forced-perspective shot. I put the white car on a tripod in front of my camera and moved it around until it appeared to be sitting on top of the table. After the fact, I did end up doing a little bit of clean-up in After Effects to remove some of the lines between the black plank on which the car sat and the table around it, but it was basically a straight forced-perspective shot.
The main problem with this approach was that in order to get a great enough depth of field to have both the car and the table in focus simultaneously I had to shoot at my camera's highest f-stop and iso setting which meant a very dark and noisy image. And even at those settings I was only able to get the car in focus at about 12 inches away from the camera, a scale at which the size gain from the forced perspective was really not dramatic at all.
This next clip shows my second approach, a kind of compromise between full in-camera forced perspective and compositing. I used a lower f-stop than last time to get a brighter image, which resulted in a shorter depth of field so I took two clips: the first one with the car in focus and the second one with the table in focus, without moving the camera in between. Then I used After Effects to glue them together:
This is nice because it lets the car be bigger and the whole image brighter, but I thought there was something not quite right about the composite. A big part was the limitations of the profile positioning. The car itself is not in enough depth. So I tried one more time with basically the same approach and a different pose:
This was the first composite that I was really happy with. The recession of the car into depth really helps sell the effect.
This trick in hand, I set out to the street to try to use it to achieve my in situ shots. The problem was just not enough light. Even at the highest f-stop and iso settings in the dark of the street there just wasn't enough light to get a solid focus on the car. Especially not close enough up in order to get it to composite at scale.
When I got back to the lab and tried and failed at these composites I decided to go back to my original approach, but with one big new advantage: since I'd shot footage of the car in position, I had a guide to as to what position I needed to put the car in for it to mate to the scene. I just setup the car in a well-lit spot and moved it around until the shadows were in the right places and it was at just the right angle.
Once I had those stills, getting them to mate into the scene completely was just a matter of getting the brightness and contrast just right, doing some mattes for things moving in front, and using AE's remap distort function to push parts of the lights in the scene onto the car (the neon storefronts, the passing car headlights, etc.).