"Consider for antialiasing" checkbox for Multimatte elements

[Pixelab]

I don't know if it's easy to implement, but somtimes when you crank down the AA to get faster results, your MM elements are (obviously) more jagged. Would it be possible to ask more sample for them (like the checkbox for VrayExtraTex)

I don't this if would add too much rendertime but it would help a lot to have clean masks with draft renders (or in general when working with higher noise thresold, even for final images)

[Morbid Angel]

you should increase your min subdivs in dmc sampler - it will be almost same as checking consider for AA.

[Pixelab]

you should increase your min subdivs in dmc sampler - it will be almost same as checking consider for AA.

You mean min subdiv in AA sampler ? I'll make a test to see the render time impact, but I don't see the point of raising the subdiv of the whole scene just for a mask.

[super gnu]

presumably if you tick "consider for aa" on a render element, it will also do whatever extra aa it does on that element, on the main render, since they are all rendered together.

i dont think you can have a crappy beauty pass and a sharp render element.

[Pixelab]

presumably if you tick "consider for aa" on a render element, it will also do whatever extra aa it does on that element, on the main render, since they are all rendered together.

i dont think you can have a crappy beauty pass and a sharp render element.

I think you can.

I should try to do it with a VRayEdgeTex in VrayExtraTex, but that's not optimal since we want antialias on contour rather than edges. A Toon shader would maybe work but since VRayToon is an effect rather than a shader, it's not a possibility.

[Pixelab]

Just made a test, VRayEdgeTex in VrayExtraTex works ( with geometry where egdes are contours) It's faster and nicer than min samples on 2.

But still it's not optimized for contours.

You cannot indeed have crappy beauty pass with sharp mask (the AA color thresold will still be the limit) but you can push the AA subdiv on geometry edges up to this thresold, whatever the sampling is on this spot of the image (eg. complete darkness will still yield a nice mask)

[super gnu]

that is true, i stand corrected!

[joconnell]

Maybe a falloff map with some very steppy transitions in the mix or output curve? It'd be hard to know which values your contours would need to be pushed though. You'd almost end up using higher aa all over for one thing, second of all since the multimatte only cares about fill and silhouette, where do contours come into it?

[Pixelab]

Maybe a falloff map with some very steppy transitions in the mix or output curve? It'd be hard to know which values your contours would need to be pushed though. You'd almost end up using higher aa all over for one thing, second of all since the multimatte only cares about fill and silhouette, where do contours come into it?

That's a really clever idea !

I've sent a test (interior with GI, 4k+) with the previous method on our farm, I'll try yours and report the result.

[joconnell]

Ooh - a handy one might be a gradient ramp with solid interpolation and a few different primary colours, set it's type to mapped and then use a falloff map as it's map source?