This has, by far, been the most challenging task as a compositor that I have ever faced. Granted, I have only been a compositor for a year or so, but I have learned more in the past two and a half weeks than I have in months about Nuke, 3D tracking, rotoscoping, proper collaborative script organization, and cleanup work in general.
Here are a few of the highlights and challenging shots I personally worked on- myself and one junior compositor were able to complete 23+ shots in under 3 weeks. Some of these shots were ten frames, but each shot brought its own unique challenges and problems to the table, and very rarely was it standard to pop down a few trackers and quickly place the wallpaper on the walls properly.
If you'll recall my last post on this film, we were unable to use a greenscreen to cover the bare walls. Bad, bad call. Dear god, bad call. The roto and luma keying was miserable and added days and days to the process. This particular shot was not awful to do, just time consuming due to the 1,000+ frames it boasts. It took me two days of rotoscoping and a day of color matching and refining to truly finish.
Here is the basic setup. This shot has a relatively simple layout- a quick garbage matte to determine the location of the edge detail key, a core matte featuring articulated roto shapes for the man moving in and out of frame, and some color grading of the wallpaper itself to match to the lighting. There is not tracking in this scene.
Here is another scene, actually the first I tackled. Since I was able to get a 3D track, I set up a projection to place the wallpaper and have it follow the highly visible trackers with ease. This scene also features articulated roto shapes and intense color matching- the wallpaper is hardly visible by the end, but this is what it would look like in this lighting
This scene in particular highlights the tracking problems we faced in many of these shots, particularly the very short ones. Because of the soft focus and steadicam jerkiness, tracking markers became impossible to use. So, in this setup, I used two camera tracker nodes, one that solves for a quick, blurry movement and one for the slow, languid movement shortly after. By switching between those nodes and two versions of the projection, I was able to Macgyver together a wallpaper that stuck to the walls.
I also made a concerted effort to keep my scripts clean and legible in this project, following a specific and personalized visual language in which A pipelines are always easy to pick out and rotos are easy and simple to find. Nodes are also labelled, particularly those that are used to create alphas for the actors.
Overall, I know there's more that can be done to further refine these shots, but I'm satisfied with how much I've learned and how hard I have worked. Yay!