Just a quick comment on this great new feature:
I noticed that it’s possible to save a rendered image with Cryptomattes as a 16-bit half float EXR, which clips the super high and low values needed for Cryptomatte to work. Most of the channels contain “inf” or “-inf” values instead of e.g. “-50023542852372352335.111”.
To avoid loads of bug reports I guess it would be best to force 32-bit full float EXR output if Cryptomatte layers are enabled. Also DWAA compression (and maybe some other types) is lossy and completely destroys the Cryptomatte layers as well. Maybe this compression type should also be disabled if Cryptomatte is enabled?