I had some hopes about karma, but… Houdini rendering engine not usable for motion design jobs

Transparent objects have full presence, what alpha should there be other than solid white?

term "transparent" can mean a few things

- traditionally it's a material with nonzero Opacity and those should be giving you appropriate Alpha channel when rendering against empty background

- but also nowadays "transparent" can relate to refractive rays, which is a bit tricky subject
if you are referring to those and that some renderers allow you to "propagate alpha" when such rays exit into an empty background or potentially reach limit then I don't think Karma supports anything like that yet

Try to render glass cube via redshift, and will see right alpha channel, I mean

Chukerman
Try to render glass cube via redshift, and will see right alpha channel, I mean

I'm not that familiar with RS, but I thought you have to turn on Refraction Affects Alpha Channel setting on for it to behave that way, which is the propagated alpha I mentioned, to fake the transparency for refractive surfaces

it will hardly be "right" or correct unless the IOR is 1
but yes, that's a pretty common hack to help with compositing if correctness is less important

in VFX workflows you would most probably want to create a pass for refracted stmap to be able to replace with anything in comp and still get distortion caused by IOR

but I agree that a robust system to propagate Alpha as well as AOVs through light paths would be very handy but also more preferable than hardcoded checkboxes for individual features so it may be worth the wait

I'd rather Houdini's native renderer not normalize such a non-physical hack that's incorrect by design.

jsmack
I'd rather Houdini's native renderer not normalize such a non-physical hack that's incorrect by design.

If implemented elegantly I'd welcome it, since it has its uses especially for thin wall objects where the non-physicality can be marginal enough to be acceptable

but more general AOV propagation through lightpaths at least refractions is very necessary
the Alpha propagation or at least mask could probably be done using LPE's if houdini LPEs supported background token B like some other renderers do

tamte

jsmack
I'd rather Houdini's native renderer not normalize such a non-physical hack that's incorrect by design.

If implemented elegantly I'd welcome it, since it has its uses especially for thin wall objects where the non-physicality can be marginal enough to be acceptable

but more general AOV propagation through lightpaths at least refractions is very necessary
the Alpha propagation or at least mask could probably be done using LPE's if houdini LPEs supported background token B like some other renderers do

As a general purpose arbitrary light propagation system, I'd also welcome it. However, I don't want to see a simple 'transparency alpha' option, as the messaging is misleading as it legitimizes using it as such.

jsmack
However, I don't want to see a simple 'transparency alpha' option, as the messaging is misleading as it legitimizes using it as such.

how about...
- a geometry property "refraction as alpha"?
- defaults to off
- when on, for paths that travel only through refractive bounces and then exit the scene, we track the final contribution, and then output that to alpha.

or does that still irk you John?

brians

jsmack
However, I don't want to see a simple 'transparency alpha' option, as the messaging is misleading as it legitimizes using it as such.

how about...
- a geometry property "refraction as alpha"?
- defaults to off
- when on, for paths that travel only through refractive bounces and then exit the scene, we track the final contribution, and then output that to alpha.

or does that still irk you John?

It doesn't bother me, but I feel like getting something that could be useful for a simple composite is more complex than tracking final contribution. It would also only be useful for parallel transparent planes where ray bending is mostly nil. I would aim for a single setting that would make the beauty pass 'compable' without artifacts such as doubled up transmission or ghosted internal reflections. How would fresnel be taken into account, or total internal reflection?

It's a feature that exists in Redshift that people have requested a few times now.
I'm no expert on its behavior or use, but my guess is they'll be using a dome light that somewhat matches the backing plate.

Haven't tried but chances are you can get it using a cryptomatte output. Should be good enough for comping when refraction is not that relevant.

Jonathan de Blok
Haven't tried but chances are you can get it using a cryptomatte output. Should be good enough for comping when refraction is not that relevant.

How would you approach this?