Hi,

Just discovered the power behind the parameters exports of shaders and the resulting extra image planes.

I have one question though, what's the easiest to composite them together in Halo ?

The network below is what I'm doing right now : I'm exporting reflection, specular, diffuse, and paint parameters. In each of the swap, I swap one of the planes with C, and then delete all extra planes to keep only the new C corresponding to one of the planes and A.
After much headache I understood that before the multiplication of paint and diffuse I had to divide them by the alpha, in order not to get a black edge around my object.
And that the composite operations shouldn't use the alpha.

I'm new to halo and compositing in general, is there an easiest way to do that ? Or a way to export parameters unpremultiplied ?

Thanks a lot,
Vincent

It sounds to me like you need to alter either your shaders to export unpremultiplied color for diffuse, specular, etc, or to change the coverage method in the “Extra Planes” section of the ROP (try minmax min). The add COPs do ignore alpha, but if it's already baked into your color, you're going to get black halos anyway. You've got the right idea using premultiply COPs to undo the baked alpha multiplication.

Thanks for the answer.
I'm using the reflective surface shader shipped with houdini, which exports unpremultiplied specular and reflection, so I thought I didn't need to divide them by the alpha right ?

But now I'm wondering if the fact that their edges are antialiased affect the compositing.

However if I change to minmax min, they are not antialiased at all, so it's quite ugly.

Hum, if anyone knows a good ressource (book or online) on these matters, I'm all ears.

Thanks a lot

IMHO the best single resource for compositing out there is The Art and Science of Digital Compositing [amazon.com] by Ron Brinkmann. Be sure to get this second edition, which is fairly new, as it's got lots of up-to-date stuff in it. It's the book we shove in the compositor's faces when we find out they don't know what they're doing.

As far as what you're doing, there's a number of different approaches, and I completely understand how things got a little confusing. The goal here is that you essentially want to combine the component parts of the final surface in the same(or very similar) way the shader does it in VOPs, that way you have the ability to override it. To simplify, though, what you want to do is examine the way the shader works, which in this case essentially adds everything together - you don't need to concern yourself with the alpha since add's don't care. The primary layers being added are diffuse, specular and reflection.

However, some trickiness ensues since you'll notice in the shader the diffuse is sent out unmultiplied by the diffuse intensity - which you'll need to take into account(or instead just wire out the clr instead of the illum and dump that into diffuseExport). Similarly, you'll notice other subtle things such as multipling the colour by the opacity before adding, etc. You'll want to mimic all this in the comp network. It seems like more work, and it is , but that's the price you pay for tweakability.

Cheers,

J.C.

Thanks for the reference and piece of advice

You should check out the Nuke forum over at vfxtalk…since this is the most capable multi plane exr compositor out there there is lot's of data that will carry over to comping in Houdini as the nodes are the same..add, mult, etc. Also working in 32 bit float linear light is important here.

This is not to say don't pick up that Brinkman book - the only real comping book that I have seen.

s

You should check out the Nuke forum over at vfxtalk…since this is the most capable multi plane exr compositor out there there is lot's of data that will carry over to comping in Houdini as the nodes are the same..add, mult, etc. Also working in 32 bit float linear light is important here.

This is not to say don't pick up that Brinkman book - the only real comping book that I have seen.

s