This topic contains 4 replies, has 0 voices, and was last updated by  Cindy Dolan 8 years, 8 months ago.

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #203131

    Hi there
    I’m doing a research paper on compositing and I’m finding that the issue of premultiplying seems to be coming up a lot. I understand that CG images can be brought into a compositing package with an integrated alpha channel and used as either straight (unpremultiplied)or premultiplied. The issue out there seems to be in making sure the system is expecting the right version of what has been rendered. If the system is not set up correctly, weird black or white fringing could result.
    I was wondering if, in the industry, it might be conceivable for artists to use separate alpha renders that are brought in independently of a primary, RGB-only image sequence. That way, with no integrated alpha channel, premult issues might not happen and time could be saved in troubleshooting comps. In AE I would have to run track mattes with each render, right? Could it be made easier if you precomp each element with its imported alpha channel? In a node-based package like Combustion, it seems like it might be easier to manage many elements like this. Does anyone out there have thoughts concerning this? I’d love to hear from industry professionals. Thanks!



    Alpha is a BW image, why would I waste the storage with RGB alphas?
    When I import 10 render passes why would I want import 10 alphas separetely?

    I work on Flint, unpremultiplying an image is just one click and there is no pain.:)

    I have no idea how the AE guys deal with this, but would be surprised if not easily.



    Cindy Dolan

    Yeah, that makes sense. The reason why I ask is that I hear of quite a few artists (proabably beginning artists) have trouble with black fringing with premultiplied (or even straight) renders when they go to comp. Also, I read that there are problems when you might go to unpremult an image (divide) in that highlights on partially transparent objects get flattened. It seems like rendering separate alpha passes might be a solution. It also might make things easier if you want to combine mattes or create edge mattes. It sounds like this is not a very common practice, judging by your candid answer, which I appreciate!:D


    Actually in Flint when you import a seq with alpha, they show on the desktop separately and you have to handle the alpha channel as an individual seq. From this point you’re right.
    As far as I know the Nuke guys can separate the alpha with one click as well.
    Since the matte is separated in the app by nature, it’s easy to work with it (combine, find edge etc.).

    The highlights are usually in separate passes and I comp them in with alpha turned off. Of course the blending mode is changed from “Blend” to something else. In general there is no premult/unpremult issue at highlights.



    I am not clear on what you are asking. There is no difference if an alpha is a separate file or contained within an RGBA.




    I think rusty is talking about the fill and not the matte. Ie is the fill rendered out of 3d package premult or straight. Rusty the matte can be seperate or not regardless of if the fill is premultiplied or not it doesn’t make a difference as everone has mentioned.

    Obviously premult creates problems with transparencies but this also includes any anti aliasing so even after you divide you can get highlight clipping errors. The best way is seperate passes all straight/unmatted.

Viewing 6 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic.