Fri, Jan 18
Sat, Jan 12
Wed, Jan 2
Mon, Dec 31
Sat, Dec 29
I assume this is similar to T57939 (in that render pipeline is not covered by DEG)
BKE_animsys_evaluate_all_animation() is already done here as an extra evaluation step.
However mask deformations are not actions, so it seems extra evaluation 'by hand' is needed... see D4140 for a potential fix...
I upgraded to the latest build (7c438e5366b2) and it works now. Thank you
Can't reproduce. Try a newer build.
I can reproduce this. When I move around the timeline, I can see the masks updating. And if I hit F12, the mask reflects what I see in the preview.
Thu, Dec 27
I think there is a general issue with save buffers and custom render engine defined passes.
Dec 19 2018
I guess I'll just make an diff
Dec 6 2018
If I change the render layer in the compositing tab to sceneB and then back to sceneA, it works as it should in 2.8.
However, if I then save the file and open it again the same problem occurs until I update the node again.
Dec 5 2018
Dec 3 2018
Dec 2 2018
This is a limitation of Cryptomatte. While it does produce accurate masks, it does not have the information to separate colors accurately. For that something like Deep EXR would be needed.
Nov 23 2018
Change the value of CMP_NODE_DISTANCE_TRANSFORM from 323 to 324. 323 was used for CMP_NODE_CRYPTOMATTE.
Nov 21 2018
assigning to @Brecht Van Lommel (brecht) then
Nov 20 2018
Technically, @Brecht Van Lommel (brecht) is the boss of render pipeline.
Can confirm this.
Nov 19 2018
Nov 17 2018
Nov 12 2018
I am utterly sorry if this was the wrong place, I didnt see it as a feature request.
My bad! Thank you for doing what you do!
That's indeed a limitation. We don't use this tracker to handle such feature or improvement requests though:
Nov 9 2018
Agree it could skip writing if the Composite node is removed (or has nothing linked), and that making it do that would be a To Do.
Agree this is a little weak.
It is not a bug though, it is still working as intended.
Not having a Compositing output Node used to complain and refuse to render, but that rule was softened in the fix for T35269.
Nov 7 2018
Working around it would be easier if the dilate/erode node had an input for step value.
I guess this is known limitation of the compositor (but I am not sure).
Workaround is (as you said) to first do the erosion first, then the scaling.
Nov 5 2018
Nov 3 2018
Rebased on the latest master.