John, was that for a star wipe?
More or less.
Yep 
It was a beast wasn’t it.
Head replace 
“Hero Fish”
Head Replace… or in modern editor terms “split screen”.
You and the video right?
lol
This shot almost broke me. Was actually 3 shots spliced together to make one long shot. Had to morph the 3 together, then replace the actress face with another take (so technically 4 shots) then do massive cg comp with fish and plants that span the 3 shots, then comp actual footage of fish we shot over the renders and finally phone replace on top of that. Took about 25 mins to render all in all and last version was v35 I believe. There are a few recursive batch setups within this setup, this being mainly the cg and fg comps. (Shoutout to Tim Farrels fluid morph. Worked great with some paint fixes - and it was a tricky one!)
Dang, all that for a source front. smh.
version 60That’s a lot of actions for a cg comp. Slow?
cool use of the compass node up at the top of the screen to breakdown the colorspace.
epic
This is a screenshot of a technical stress test from 9 years ago, to the day.
It was designed to prove that memory leaks and render errors were caused by Debayering RAW files, but those issues would diminish or disappear if the sources were transcoded to uncompressed intermediate files.
The batch contains 512 unique 600 frame sequences, each of which was 6144x3160@16-bit uncompressed openEXR files, transcoded from an early model Red Dragon.
There were 307,200 frames.
Each frame of each sequence weighed ~110 Megabytes.
The total dataset for this batch was just north of 33TB.
This prevented any file system caching.
512 layers weighs about 55GB for each frame of the batch render.
This prevented any GPU caching (NVIDIA P6000 24GB).
The combined horizontal resolution before final comp was 199,608 pixels.
The combined vertical resolution before final comp was 50,560 pixels.
(This is about 9,940 Megapixels)
Each composited frame took 45 seconds to render.
There were no errors.
For comparison, 128 layers of R3D raw files crashed the render after 30 frames and decompression errors appeared on random layers at random times - totally unusable.
(I would have transcoded to 32-bit files for more precision but I had limited space for the dataset(
I think you wanted no one to be able to revise.