Cryptomattes implementation
Psyop is a very interesting and talented studio and they have managed to develop this method that combines Z-depth and ID render passes to generate a custom render layer that gives the ability to deduce a mask of any object in frame, while in comp!
psyop.com/news/view/new-cryptomatte-tools-stretch-beyond-psyop
It would be interesting to have a blender implementation or similar approach
11 Comments
They're being developped. Watch this .
@Caetano Veyssières: Great! I han'd seen this. Thanks Caetano
@lozo.vfx: yeah you could ask the tangent dudes they might be able to have a build with it that they can share, I think they're working together with the BF to develop and implement this for 2.8
@Luciano Muñoz: Well it would be great even if I'm affraid that it may take longer ... but we'll see. Finger crossed
I believe that its possible in Cycles and also with thw new feature on the viewport.
@Charles Lol: that's what i'm thinking as well but the differences happen to be the selection methods as to how you select what in blender's compositor. if all you want is the mask i think that a mix of material id, object id, and zdepth would do this.
if you want the "picker" style mask selection it's not so readily available with what we've got now i think.
There is a git branch here: git.blender.org/gitweb/gitweb.cgi/blender.git/shortlog/refs/heads/temp_cryptomatte and a thread on BA here: blenderartists.org/t/cryptomatte-1-0-released/671348
Also, let's upvote this! :)
How would you envision the Blender implementation working? Would this just be a separate pass that users would enable on a render layer, or is there something more involved that you expect they would use? Could we still use the ID Mask node to pick objects... or would you expect users to make these selections based on a color picker?
@Jason van Gumster:
Cryptomatte's approach is made of tow steps, one is during render time writing the necessary data in a render layer/AOV and the other is through a comp node that reads this data to generate the mask through a picker, a material name or an object name.
@lozo.vfx: So like object and material indices, it's a pass in the render layer that users enable. Would you expect that this feature would replace those? And as far the compositor node for generating the mask, how would you expect Blender's it to look in Blender? I'd assume a single input for the pass data and a single output for the mask. How would users choose between the picker or name-based interfaces?
(in case it's not clear, I'm looking for mock-ups here)