Cryptomatte in fusion - anyone else finding it unbearably slow?

[Neilg]

Testing a new workflow for our next animation project. I like the idea of cryptomatte - we have lots of shots of similar areas with consistent lighting so using a single loader would be hugely beneficial, but I just converted a 4k shot that previously used 10 multimattes to use crytomatte instead and my per frame render time jumped from 4 minutes to 25m.

This is frankly terrifying when we've got 9 minutes of animation to create -

Render started at Tue 12:38PM (Range: 0.0 to 0.0)
Render completed successfully at Tue 1:03PM - Total Time: 0h 25m 27.87s, Average: 1527.87 seconds/frame

anyone else finding similar issues, or have tips on using it?
I read on a forum someone saying they cache it all to disk, but they were complaining of a per frame increase from 1 to 5 second so it hardly compares. plus that's a pretty awkward process in fusion and updating multimattes the old way is just as much work.
Anyone else finding ridiculous render times with it or am I doing something wrong?

[joconnell]

We found the same thing in nuke on huge scenes - cryptomatte apparently stores values in a small amount of huge dictionaries so if you've tens of thousands of objects in your scenes, it's very slow to process the operators after.

[Neilg]

That'll be it then. this test scene has a ridiculously high number of objects, but mirrors the workflow we are using for the new project. Doesn't sound like it can be worked around very well...

We'll go back to object and material channels, just need to get a script which can replace 20 loaders automatically.
I really wish you could do the channel / format stage after the loader in fusion so we can use wireless links. Re-linking every channel manually even if it's an embedded exr seems to miss the point of them.

[joconnell]

Yeah I don't know if anyone's actively developing or maintaining it, if it's open source perhaps we could have the project added in to a new open source foundation and have it improved...

[Neilg]

There was an update not too long ago - https://github.com/Psyop/Cryptomatte

I was planning to request a feature - if you have the cryptomatte levels set too low in vray, it crashes in fusion. Need some way of knowing how many levels you need to use based on the scene, or a warning that the levels are too low in the plugin.
The more I think about it the more it seems like it'll never be a quick option - it needs at least 4, 32bit r g b a channels to work so it can store text within the massive rgb values. That's never going to be fast to read is it...

[cheerioboy]

Wonder if it matters whether you have it set to objects vs materials - might end up being less data if by material? I've switched to having the cryptomatte work via materials and the only snag is that multi-materials need to be exploded. But it's been helpful in that I'm not constantly trying to chase down what channels have been used already for multimatteelements. I'm only working with still images so I don't have enough experience with cryptomatte & animation work..

[Ihno]

I've heared that those lua scripts are a bit slow in general.
The Fusion implementation is not developed by Psyop as you may already know. Its a community project.
https://www.steakunderwater.com/wesu...php?f=6&t=1027
I'm afraid it would be up to Black Magic to make a proper and efficient implimentation.

I'm now using the slplit into multipart exr option in the option re. Because the way you choose channels in fusion is way better than. And it loads so much faster.
But I can't get Cryptomatte to work with that.
I also went back to Multimatte for now.

[squintnic]

agree only useful in small scenes, no good in large arch viz scenes.
wasnt sure if it was fusion and not tried in nuke

unimpressed with fusions dev since blackmagic took over, feels like its dying out. support is atrocious. reckon it will be wound up soon now the resolve 'integration' is there.

[Neilg]

unimpressed with fusions dev since blackmagic took over, feels like its dying out.

would have been dead by now if they hadn't anyway, not sure who else could have taken it. chaos groups fusion would've been fun...
So long as they keep it stable and maybe make third party scripts and plugins more efficient i'll keep it around. I get the impression the dev team is pretty small.
Is it possible to open existing fusion files from within resolve?


I wish resolve was stable. nobody in this studio can use it without it crashing constantly. boots us to desktop when doing basic shit like adding clips, trimming and scrubbing, the only software i've used that crashes more frequently is vue.
I was considering emailing them to ask what hardware we're supposed to use to make their software work, but also based on principle (it should work) I probably wont.

[squintnic]

We moved back to Premiere after 6 months battling with Resolve and its stupid database!