GI Animation flickering please HELP!

[^Lele^]

Full brute force, maybe even with only 2 bounces (outdoors require less, complex objects like cars and so on, when moblurred might look allright with 2 or 3 bounces).
Using Adaptive DMC as AA type will insure rendertimes won't skyrocket.
Using darker materials and proper camera exposure (similar to what outlined in my old tutorial on sun and sky) will in this case lead to much faster rendertimes, along with a nice and contrasted look.

[andrewjohn81]

I'm not sure that will work well for what I've got though.

I have a wall of plain white cabinets, no trim, no handles, and a fairly basic white or stainless appliance as the focus. There's no environment other than that, a vray plane for the floor and a plane for the wall. that's it. So it's kind of an outdoor scene as far as geometry is concerned because it's similar to looking at a building, If it were plain white with a bunch of moving parts.

I tried DMC, but I just can not get rid of the noise. I had to raise the Brute Force GI subdivs to at least 24. That's where the noise is getting acceptable. And that's with the Noise threshold at .001. A very small portion of just the white cabinets takes 1.5 minutes and it only about 100x75 pixels in size. I'm doing an odd size animation at 700x700, so that would take too long I think.

Any suggestions on how to render a scene like that out? The designers always want their product in a very basic environment, plain white, so it's not distracting from the product at all. Makes it hard to do an animation with lighting that they like.

[^Lele^]

I would suggest you to try and raise BF subdivs without touching the noise threshold, particularly with schlick.
It's going to be slower than an interpolated method, that's for sure.
Get to 32 or 48, even.
And then use the AA only for the edges and texture, rather than have it help much on the GI phase (a la Universal Settings).
This way, you should be able to decouple GI calculations and AA (as the GI cleans "before" the AA).

[andrewjohn81]

I'm doing a test of about 2% of the image and it's taking about 2 minutes, and that's on pretty much the most basic part of the scene. I'm just testing to get rid of the noise. I'm guessing 2hours+ for the renders then. I think that's a little long. I'm hoping to find another way.

Should I not be using Brute force for both Primary and Secondary bounces? I wonder if I would get good enough results with it only in Primary and just raise the multiplier to the value matches pretty close. Although, I assume that would cause problems because if I do that it would ignore things that aren't visible to the camera, like the underside of the cabinets right? I'm not sure if my understanding of Primary/Secondary is correct. Is primary anything that is in Front of the camera, or only rays that hit an object from the camera's view? If it's any ray that would be in front of the camera then I think I could get away with just Primary for this scene.

I'm having to raise subdivs to at least 64. I couldn't see a difference between 96 and 128, but i can see the difference between 64 and 96, and there is a significant time difference. I think 64 will work out, but it still really slow. I'm just hoping to speed this up now.

I think having some noise is better than having flicker though. I can always reduce a little noise in post, to a degree, without losing much info and at minimal time.


btw, I appreciate the help. I'll be going home soon though. the snow is getting pretty ridiculous here and people are starting to leave early. I can't believe it. Snow this late in the year!

[^Lele^]

I'm rendering a little test right now (small 320x240, 3 balls bouncing, similar setup to yours).
I'm using only the primary bounce for it, so it's similar to an IBL rendering.
Primary only will light the underside of objects too, consider it the first bounce for GI (the very first bounce, direct lighting, you can consider as bounce "0").

[andrewjohn81]

well I submitted it at 64 Subdivs for all the products. It adds up to about 1000 frames, which I estimate will take about 6.4 days of rendertime on our machines.
Using this method it shouldn't matter what machine i render on right?
I noticed with LC and IRmap if I used one machine with a different configuration it came up with different results.

I can add in 4 more machines it the results don't change.
Noise isn't horrible, but flicker makes you feel sick to your stomach.
We'll see some results when I come back over the weekend.


I'd like to see what results you come up with for the ball test. If that works out well I'll try that with future jobs when we need white scenes in a hurry. I'll try and push darker background scenes more often. lol

[^Lele^]

So the settings were as described (32 subdivs, Brute force GI, first bounce only, 16 subdivs Area Light, 1-4 DMC AA with area filtering).
Notice the scene is terribly simple, but I had a box pass over the light and block it for a few frames (which would likely result in streaked GI with IRMap set to animation mode).
Noise is barely visible at 200% zoom, and only if you go look for it.

http://www.megaupload.com/?d=1ENBCPRZ

Here's the 5 megs rar containing an uncompressed AVI for you to play frame by frame.
I appreciate things will get more complicated with complex shaders, however, the GI will not incur in substantial speed penalties, with only one bounce.
Notice that applying a gamma ramp (or curve color correction) to this will brighten the shadows and fake a more complete GI solution with a fraction of the cost.
Of course with one bounce only you'll lose most of the color bleeding.
However, adding one or two secondary bounces will not impact time enormously (mainly only double the GI part of the calculation: 1:44 for one bounce, 3:10 for two, 3:59 for 3 and so on.).
More than one secondary bounce in such a setup will not lead to significant differences, though, as the times testify.

[andrewjohn81]

well, most of my renders finished. I'm still waiting on 2 sets which should finish within the next couple days. I've got some stainless steel and the noise animates quite a bit, but I can fix that in post for the most part.
Times are between 1 and 4 hours though depending on what's opened.

I tried some test using one bounce on the secondary, trying to remove the secondary completely, but they came out Way too noisy. I couldn't get the noise off the floor at all.

So is this how everyone is animating when they have a highly animated scene? That just seems so long.

I haven't had a chance to look at the scene you made, Lele, because our IT department blocks that site. i'll have to check it out at home. They pretty much block all the major file serving sites because they see it as being primarily for illegal purposes. I sent them an angry email about it.