Is it possible to render without premultiplication in mantra? I can't seem to find the switch…
Thanks,
MD

Not as far as I know. You can post-process it though.

wow, if that is the case, that is a major oversight. I know you can unpremultiply, but then you end up quanitizing your 8bit alpha to 7bit by the divide, which can bite you later.
I find it hard to believe, I see that the ‘sample filter’ is set to alpha according to the console, so maybe it's somewhere there?
MD

Huh? Where have you seen a visible problem if you're handling your compositing correctly(convert to float first - or better - just be working in float by default)? I don't understand what you're referring to.

Are you talking about a theoretical issue or real-world?

Cheers,

J.C.

a few real world scenarios:
I don't the luxury of comping the best tools out there, and often it is other people who will be doing the comp. After effects will allow you to work with premultiplied images, but it exhibits the truncation I descibed. Usually not a big issue, but dense thin lines can exhibit issues, like fur for example. Perhaps running AE in 16bit mode or even float will solve that, but I don't really give them the benefit of the doubt of whether they perform the unpremultiply on the 8bit image or after it has been extrapolated to 16bit. A worthwile test, but I have never need to worry about it since I work with straight alphas in other packages.
Second is the Flame, which the operator has to manually unpremultiply the image, and usually they need reminding, which I would rather not do. Again the bit depth is an issue, since flame is not too graceful in handling multi bit depth images
I cannot get anyone here to switch to Shake since apple end of lifed the current incarnation, with no news of the next iteration. I wish I could go fusion, but our guys are on macs and flames.

anyway, sample filter does not perform the function either. I will try writing out a file with just the Cf plane next

MD

mattd
a few real world scenarios:
I don't the luxury of comping the best tools out there, and often it is other people who will be doing the comp. After effects will allow you to work with premultiplied images, but it exhibits the truncation I descibed. Usually not a big issue, but dense thin lines can exhibit issues, like fur for example. Perhaps running AE in 16bit mode or even float will solve that, but I don't really give them the benefit of the doubt of whether they perform the unpremultiply on the 8bit image or after it has been extrapolated to 16bit. A worthwile test, but I have never need to worry about it since I work with straight alphas in other packages.
Second is the Flame, which the operator has to manually unpremultiply the image, and usually they need reminding, which I would rather not do. Again the bit depth is an issue, since flame is not too graceful in handling multi bit depth images
I cannot get anyone here to switch to Shake since apple end of lifed the current incarnation, with no news of the next iteration. I wish I could go fusion, but our guys are on macs and flames.

anyway, sample filter does not perform the function either. I will try writing out a file with just the Cf plane next

MD

If you're rendering from Houdini, it shouldn't be that tough to render to 16 bit int/float (or 32 bit float), do the un-premultiply then quantize down to 8 bits for AE.

If you're concerned about lack of control, you should definitely be rendering to float or at least half float, and prep/precomp your layers. Float compositing is (slowly in some corners - *cough*Discreet) becoming standard nowadays. Even for 8 bit targets I never render less than OExr at half float nowadays - and we have three Inferno suites here. Btw, you can quantize your layers down to 16 bit integer tiffs and Discreet can input and re-quantize that yet again to their proprietary 12 bit int depth. It's not perfect, but it sure beats 8 bit.

Cheers,

J.C.

You also need to watch out if you render with anything other than gamma 1. The alpha channel doesn't get gamma corrected so the C and A can get out of sync. Not sure if that is an issue with rendering floats, but we render straight to 8 bit gamma 1.4 normally, and I only found out very resently that the gamma channel wasn't getting changed.