aiFlakes are swimming during animation

[rmelville]

I'm working on a product shot that involves a car paint material with flakes, and I can't get the flakes to stick when the object is animated. They float around and create a ton of noise. I'm not sure if this is just the nature of the aiFlakes node or if there's a way to address this.

I've got my aiFlakes node plugged into a standard shader normal cam slot. I've tried using different iterations of world/obj/pref/uv coord space, in addition to using the aiTriPlanar node for projections and none seem to resolve the issue. I've also tried increasing cam/diff/spec samples and locking sampling pattern in the Arnold globals. Locking sampling pattern seems to help resolve inconsistencies between frames while the object is static, but the flakes swim around once the object starts moving again.

Has anyone run into this issue?

p.s. I should also say that the geo is an imported step file converted to poly, with auto UVs set up in maya. I'm using Arnold 5 w/ Maya 18

[rmelville]

So after a bunch of troubleshooting, it looks like the problem was in fact under-sampling. I'm using very fine and dense flakes, and the resulting noise in the renders made the flakes appear to swim. I really cranked on the Camera samples (increased to 10), and now the flakes are crisp and clean with minimal jitter. There may be a more efficient solution out there, but for the time being, this is my best fix. I'm pretty new to Arnold, so if anyone has other suggestions, let me know. Thanks!

[maxtarpini]

The problem actually still _is_ there. You may have it fixed for that particular scene or animation but it will pop-up as soon you animate the camera moving close or far from your model for example or if you'll animate your lights making narrow angles with your models. And sadly even without looking at any image I can say that no filter strategy can fix the problem nor you can fix it by increasing the sampling settings.. in facts the problem is not about 'under sampling'. The problem is that aiFlakes is a naive implementation of flakes based on procedural noise so it cannot simply account for features that take up a very little pixel space. It's not about Arnold, it's about the classic (Monte Carlo) pixel sampling.. you can shot as many samples you want that you'll never ever catch consistently those minuscule features. The whole problem can only be solved with an ad-hoc implementation that starts from those tiny features and sample up the pixel not viceversa.