I keep hearing this has been fixed in 5.2, but that remains to be seen for me. I even saw the bug report online, but I think it was primarily aimed at cloth. They said it was fixed, but its not. Cloth still doesn’t seem to behave correctly when I use Temporal AA on MRQ. Not only that, ANYTHING with physics behaves unexpectedly. Hair, Cloth, Niagara Particles, you name it.
For a shot, I have a subtle Niagara cloud being emitted that floats, but needs to inherit velocity from character movement. Looks fine in engine and when simulating. But once you turn on Temporal AA in MRQ, its like the inherited velocity is being being insanely beefed up at like 10x. And the particle is literally flying out of the scene.
The developers on the bug report mentioned its because time dilation is bugged when using Temporal AA in MRQ? But they mainly tried to address it with Cloth. But why only cloth when many other physics related features suffer as well? In general, why is this not being addressed or taken seriously? You guys are developing feature after feature but are breaking essential systems that render these features useless. There is a much deeper issue going on here. Its not just Cloth. Anything with physics gets messed up by Temporal AA. So much so its just not useable.
I think OP is correct, and that it is also a motion blur issue. There are some work-arounds but with some pretty bad consequences. So if you have anything with physics in your scene, you either…
1: Turn motion blur all the way up to 1. But you get a very milky look in the render that isn’t desirable for cinematic production. Motion blur looks much more desirable for me at 0.2.
OR
2: You keep your Temporal AA sample at 1, and sacrifice higher quality edges, detail, dof and motion blur.
Both are difficult decisions to make with very unfortunate compromises.