-
Bizzare issue with AE handling alpha channels
I came across a seemingly strange bug in AE the other day, and I thought I’d throw it out to the COW to see if anyone had seen something similar. This came about when I was speaking to a friend who’s a Flame Artist, and he told me he had problems pulling matte’s for images that were generated from AE as 24-bit RGB images with separate 8-bit images rendered for the alpha. He said he noticed a distinct difference when using a sequence that came rendered from AE as a 32-bit sequence with an alpha versus having a a sequence rendered from AE as the fill and then a separate alpha sequence rendered for the matte. Using a 32-bit sequence (premultuplied alpha) produced clean text with no jaggies, while combining the 24-bit fill with the corresponding 8-bit matte produced jaggy text that appeared to be slighly choked. At first I just assumed that something was probably prepped for him incorrectly, but I created a test project here off of a recent animation we were working on and I found the exact same issue. I was wondering if anyone else has seen this problem. The simplest test is to create a text layer with a drop shadow and render it out to a 32-bit file format. Then take the same text and render it out 24-bit RGB (black background) and then render a second image 8-bit alpha only. Create two different comps with the same background image. In one of them place the 32-bit text render and in the other place the fill and matte renders and either use a track matte on the fill layer or use the set channel filter to define the matte. Do you see a difference between the two files? Technically speaking these should both be identical, however, I am finding different results. I am posting a sample of my findings below and I’m curious if anyone can explain why this might be occurring or if it is in fact a bug. Really just some food for thought as I like to know several different means to the same end whenever possible. Curious as to your findings. Here is the sample image:
https://www.edit1.tv/clientaccess/mpeg/capital/Text_problem.jpg
-
5 Replies
-
Hi Michael, render your 24 bit file out straight not premultiplied (which is the default) and it will work fine.
Good luck
Pete O’Connell -
Hi again, I just though of another way. Precompose the 24 bit image and the alpha layer luma matted above it, then in the main comp apply the effect: channel>Remove colour matting, to the comp containing these 2 layers.
Hope this helps
Bye
Pete O’Connell -
No, your problem is none. Like Pete said, you need to work with straight Alpha. Internally, AE uses that all the time as does the flame*. In your example, you are simply degenerating the Alpha by imposing an additional level of transparency with your matte onto pixels that already have the correct transparency, but combined with the BG color. It is in that case inevitable that the results are wrong and the edges look jaggy – if the opacity of a pixel falls below a certain value, it will effectively become invisible and only the strong ones remain.
Mylenium
[Pour Myl
-
Thanks guys – I appreciate both of your replies. Just out of curiousity, in what scenario would you ever want to use a 24-bit pre-multiplied RGB image? It seems that you would always want to be rendering straight RGB with a corresponding alpha if you want the correct results.
-
As a general rule – render straight. However, some prgrams may not like a straight render (I;ve heard this, but necver actually seen this issue) so in that case, you might wnat to try it premultiplied. But like I said, Straight is the way to go.
If you are not up on AE straight Vs. Premultiplied check this tutorial out:
—————————————-
Aharon Rabinowitz
aharon(AT)yahoo(DOT)com
http://www.allbetsareoff.com—————————————-
Click the link below to subscribe to the Creative Cow After Effects Podcast, and get free AE video tutorials:https://phobos.apple.com/WebObjects/MZStore.woa/wa/viewPodcast?id=111087911
Reply to this Discussion! Login or Sign Up