This looks really nice at first, but I'm kind of getting the feeling, that this is more of an awkward workaround, than a real implementation of cryptomatte.
First of all, even after downloading the latest version, my cryptomatte passes don't get imported the same way they look in the tutorial and for some reason don't seem to allow me to do flood selections (probably because the values in the cryptomatte pass aren't really colors). My suspicion is, that before comp, they saved the cryptomatte, that is supposed to be an EXR to work properly, as a tiff or png or something from the vray frame buffer (so pretty much just a diagnostic view of the cryptomatte instead of the cryptomatte itself), to get the color info into affinity photo, discarding the data that makes the cryptomatte pass actually useful in the process. You can see at the file tab, that it's a 8-bit file, that could never hold the necessary amount of data for an actual cryptomatte layer to work. The workflow shown here looks more like using render ID passes, which I've seen some people do before cryptomatte became a thing, but's not the same. Please don't get me wrong, this can save you a lot of time setting up masks, but doesn't give you the functionality cryptomatte was supposed to give you.
The way cryptomatte is supposed to work, is, as far as I understand it, that you get a channel for every object/material/tag, that the cryptomatte pass is supposed to capture and you end up with a file, that basically contains an alpha mask for every unique tag in the scene. Sure, the files get huge, but you get really nice selections including antialiasing and sub-pixel details.
Kudos on the spare channels trick though. That is something I didn't know about and it looks super useful!
But back to the topic. There's two problems with the workflow presented here:
+one is, you're limited to 100 objects, that can be stored in the mask, if the Flood select tool slider works with 1% increments, which can be a problem in more complex scenes with 100+ tags, maybe even less, since the colors are assigned randomly and can be pretty close together at times
+the other, much bigger problem, in my opinion, is the selection borders. Since you're not really extracting cryptomatte data, but only doing a flood select on a flat color image, you don't get antialiasing working and you lose definition in areas with very fine detail, often in sub-pixel scales. You'll also probably get nasty selection borders with glowing edges or some kind of ghosting, if you try and do some kind of a more extreme adjustment.
Maybe I'm just clumsy, couldn't get it to work and am misunderstanding something, but I find this video a bit misleading to present it as working with cryptomattes.