multimatte and vrayblend 2 help?!

[super gnu]

ok so further to my frustrating switch from gpu to cpu to allow me to use multimatte with vrayblendmtl, ive found out that basically, it doesnt work in cpu either.


it works insofar as i get a matte, but it doesnt follow edge of material at all.. i get a nasty messy edge


the mask for the material is a high res, sharp, b/w jpeg loaded with hdriloader.

the material looks fine.


if i change filtering mode, the messy edge in the matte changes significantly, but doesnt get fixed.

changing to normal bitmap loader with normal filtering improves things, but still the multimatte is nowhere near the material edge.


im guessing the fact that in some pixels the material is blending with another material, due to antialiasing of the mask, its confusing the multimatte..

what i do know is ive got a deadline and i cant get mattes for my materials... client is getting antsy.

heeelp?

[super gnu]

ok vraymtlID doenst work either, and vraymtlselect has a correct edge, but shows refractive stuff on top of the selected material, instead of being refracted only, like multimatte

[super gnu]

any alternatives, ive tried everything i can think of.

if the only alternative is doing a pile of separate renders im going to be screwed.

[kosso_olli]

Can you show a tiny preview of what the edges of your masks look like now? I have an assumption of the cause for this...

[super gnu]

sure..







first is the multimatte, note the crappy edges.. second is a mtlselect... no good cos it contains the droplets, not just refraction distortion, third is a mtlid, which shows same crapiness as multimatte, and final one is the render.



you might need to zoom in to see the issues clearly.....

it seems to me, , these mattes have no idea what to do when a material is masked by an antialiased map.. the pixels are probably not 100% that material near the edges or something...? i class this as a bug, since if multimatte can deal with antialiased rendering, where two materials occupy the same pixel, then they should be able to deal with this.

[super gnu]

ok quick update, i appear to have found an unlikely workaround.... waiting to see what the catch is.

if i enable "incompatible" materials, and use 3dsmax blend material to mask one material from another, the multimatte gives a nice correct edge.

obviously for the 6 materials of the can im going to have to stack a load of max blends up, and im sure its going to bite me on the arse somewhere else, it wouldnt be "incompatible" otherwise....

however, shows its definitely a bug and can be made to work.

[super gnu]

ok. the catch is it doubles my rendertime. great.

[kosso_olli]

Well, the preview is very small. The scene itself is not overexposed, right?
Also, your bitmap blur is set to 0.001 for the blend mask?
What you are trying to achieve should be easily possible, we have done this a dozen of times before. In general Multimattes are very robust.

[super gnu]

ok, so going back the the vray blend mtl and setting filtering to 0.01 (as low as it goes) i get a pretty good result, not perfect, there is some tiny amount of fringing, but probably good enough.

strange that max blend mtl works ok with filtering, but vrayblend does not.

i didnt try with no filtering before, since all the advice ive read is turning filtering off leads to higher rendertimes, and is not advised.

[kosso_olli]

i didnt try with no filtering before, since all the advice ive read is turning filtering off leads to higher rendertimes, and is not advised.

It may increase render times slightly, but for print work I accept that instead of having the mushy and blurry textures you get with blur of 1.

[super gnu]

ok so now im doing a massive high res render, and im back to having problems with blendmtl masks, and multimatte.


see attached, render, matte (note crusty edge) and result of using matte to change colour of render.


this is with bitmap filtering of mask jpeg set to 0.01

seemed to be ok for an HD frame, but now im at 10k res, its playing up again....

[kosso_olli]

You have set the dropelts set to affect all channels for refraction, right? How are they done? With the old displacement-trick or are the scattered?
Also, what are the min and max values for the AA?

Edit: I just loaded the render and the mask into PS. For some strange reason, the green area of the mask is much much smaller then the white triangle in the actual render. I have no idea where this is coming from. It looks like the mask is kind of refracted or shifted or something. The droplets line up perfetly, though.

[vlado]

The issue happens because the masks are in JPG format and they have some compression artifacts. To fix them, you can use Color Map feature in the Bitmap textures like in the attached screengrabs to scale the masks a bit below 0.0 and above 1.0. This will make the artifacts to go away.

[kosso_olli]

Ah, okay. Now thats an easy fix.

[super gnu]

just to note, those jpeg masks were saved out of photoshop at 100% compression quality.. so seems a good reason to avoid compressed maps for now.