Re: Alphas, mattes, adds and overs

Date : Tue, 26 Jun 2007 14:56:47 -0400
To : XSI(at)Softimage.COM
From : "Ponthieux, Joey" <j.ponthieux(at)nasa.gov>
Subject : Re: Alphas, mattes, adds and overs
After looking at your scene what I see here makes perfect sense. A premultiplied image is optimized for an "over" like composite, in other words its expecting to be composited over something else, not joined or mixed. You are attempting to create a boolean of object A to object B, this is why ADD works and OVER doesn't. The object edge which has background color mixed into it will not join with the second object which also has background color mixed in.

Using OVER one object is being placed over the second which means that the other object is acting like a background, in other words it doesn't care what the alpha is for the second object. Since Over is optimizing the top layer with the background(second object) it is more or less dumb and not getting any premultiplication treatment that you would expect from the second object, so the black and transparency in the edge is finding its way through the top layer from the background(second object) layer.

Using ADD, you are technically mixing the two where there is color or value. Color and alpha from one is being added to Color and alpha from the other. Where there is no color nothing is added. (I assume you are using Math Composite/Math op/Add in FXTree). Add really isn't a composite tool like Over as much as it is a "joining" or boolean tool. Adding the RGB as well as the alpha is probably not a problem as long as you are adding the two objects together over black and output that result for later use, any other circumstance will more than likely yield unwanted results. This result will then need to be comped OVER the rest of your visuals. This is easy to do with Shake, FXTree etc, since you can isolate a node structure separately, but probably a real pain in AE since it is layer based and is wanting to add everything from lower layers to the top layer(as you noted). If there is a way to add the two at the bottom of an AE project then may you can OVER them later, I don't know, I use Shake here and though I have AE I rarely use it.

In any case your original assumption is correct, Over will not yield the results you want and Add, while probably not the only method sure looks like the best one for this problem.

Joey Ponthieux
NCI Information Systems Inc.
NASA Langley Research Center
____________________________________________________________
Opinions stated here-in are strictly those of the author and
do not represent the opinions of NASA or any other party.



adrian wrote:
we're rendering premultiplied, and the result is fine, as alan says, if we composit using node based ADD, then your background gets dropped
in later down the tree


however, lots of people still comp in afterfx (cos it's better damnit, don't argue!)
the issue then is when you ADD on layer over another, any layers UNDERNEATH are also added to, and you get gohstly glowing 3d over
your backgrounds..... not cool!


i have put a scn file here, if anyone wants repro;

http://www.fluid-pictures.com/public/matte_conundrum.scn

a

Adrian Wyer
Fluid Pictures
33 Glasshouse Street
London
W1B 5DG
T: +44 (0) 207 434 5133
www.fluid-pictures.com
----- Original Message ----- From: "takita" <takita(at)earthlink.net>
To: <XSI(at)Softimage.COM>
Sent: Tuesday, June 26, 2007 4:49 PM
Subject: Re: Alphas, mattes, adds and overs


Are you rendering unmultiplied (e.g. what AE folks call "straight") out of CG?

What might work (dodgy sometimes) is to render premultiplied and run it through the unmult op in FXtree or AE's unumult plugin (lame, I know...).

But sometimes this helps make the edge consistent at least.

-T

--


A.K.I.T.A.: Artificial Killing and Immediate Troubleshooting Android

---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi



---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi


---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi

  • Follow-Ups:
    • XSI Viewer
      • From: jentzen mooney <jentzenskills(at)yahoo.com>

Search the XSI List archives here or use the advanced search form to search across mailing lists. Searching help is available.
This site supposedly brought to you by Benjamin Grosser and the Imaging Technology Group.