Fixing the Render Pipeline: Animation Capstone 4
Last year, I was the Lighting & Rendering lead for the Animation Capstone. Our rendering pipeline had significant organizational issues, which left us with a lot of beautiful images finished months later than we needed them. This year I am TAing for the Capstone, and I intend to fix that.
There were two major problems that we encountered:
Creating cohesive lighting between 10 different lighters working on 100+ shots was incredibly difficult, time consuming, and obnoxious.
Iterating on general feedback across the whole film and broad asset updates was incredibly slow. An update to the ground textures took months to spread through the film, and overall tone shifts were enormous undertakings.
Our lighting pipeline was to assign each individual shot to a lighter, where they would create bespoke lighting setups for that particular shot. We did our best to assign shots in chunks to make the lighting consistent between shots, but there were inevitably sequences split between multiple lighters, and recurring locations that were lit by different people. This lead to a long period of tweaking shots to match each other.
Our rendering pipeline involved using a tool from the shot file to send the individual shot to the render farm, followed by manually compositing the resulting images into a shot file in After Effects. This had to be done manually for each individual shots, and as the film progressed and scenes became heavier, long loading times made this even more painful.
With this year’s film, I am reorganizing our pipeline in a number of ways to solve those problems and bring us closer to industry best practices. For one, we are pushing back when we start lighting, so that we can start lighting with mostly complete assets and minimize repeated work. Also, we are building base lighting setups for each location and time of day, which can then be referenced into the actual shot files. This means that every shot will use the same base lighting, and that the lighting can be edited for an entire sequence all at once. To cap it off, I am building a tool that can automatically send every shot to the render farm, and another tool to automatically composite the latest renders for the whole film.
Between better organization, updated techniques, and more powerful tools, I hope that we can significantly less work overall, and iterate an order of magnitude faster. Instead of having to update 100+ shot files, we can tweak ~12 lighting setups. Rendering and compositing the whole film, which currently takes several days of active grunt work, will be as simple as pressing two buttons.
This workflow is still untested for our particular group, and I won’t really know whether it will work for another few months, at which point we’ll either have a completely rendered film or a set of new and interesting problems to deal with.