6 Replies Latest reply on Oct 26, 2010 3:56 PM by Navarro Parker

    How do you get 32-bit float effects to work with an alpha channel?

    Clint Porter Level 1

      I make a few bugs and lower thirds, but I'm having a lot of trouble getting 32-bit float effects to correctly export with transparency.


      When I make semitransparent animations with bright lighting effects, they look great in AE but I can't seem to find any way to bring them over to Final Cut and get them to look like they should.


      I've attached this example of a bug with a screenshot from AE on the left and from FCP on the right. The colors dim and the vibrant blue gradients get flattened into a dull greenish color.


      Screen shot 2010-10-18 at 3.51.04 PM.png


      So far as I can tell, all my alpha and color management settings are correct. I've tried endless settings options and codec options to try to fix this, to no avail.


      You can create an extreme case as an example by creating a semitransparent shape, cranking the color way up to a saturated superwhite color, and moving it around with motion blur. The render looks pretty much just like a 8/16 bit render, and nothing like it did in your canvas.


      I have a feeling the way AE multiplies the superwhite color values with the alpha channel gets "lost in translation" when it is rendered to 4 8-bit channels. Is there any way to get it to look the way it does in the canvas?




        • 1. Re: How do you get 32-bit float effects to work with an alpha channel?
          Clint Porter Level 1

          An explanation of the example might

          be helpful:


          1. bug during 32-bit reveal

          2. detail of the first image

          3. bug resolved, but with slight 32-bit effects applied

          4. project/export settings used

          • 2. Re: How do you get 32-bit float effects to work with an alpha channel?
            westonwoodbury Level 1

            Try bringing the export back into AE or another program, is it faded there?  That way maybe you can tell if it's FCP/QT, or AE's issue (if it comes back into AE faded, yeah maybe there's something wrong within the export settings).


            I've had big problems with QT on OS X displaying colorspace, brightness, etc, however it wants and inconsistent with other software.  One thing you could try is using the "None" codec instead of "Animation".  I've had to use it a few times when Animation was doing strange things for me.


            Another thing to look at is the Color Management tab on the output module from the render queue.  Experiment with 'Preserve RGB' and "Convert to Linear Light', as I believe these have to do with how it handles super whites and such--though I'm not 100% sure on that.


            - wb

            • 3. Re: How do you get 32-bit float effects to work with an alpha channel?
              Clint Porter Level 1

              Thanks so much for your ideas. I'd already tried experimenting with a number of color management options, including 'Preserve RGB' and turning off 'Convert to Linear Light' but they didn't fix it. I had also already tried using the "None" option from the codec menu.


              I pulled it back into AE, and it looks the same as it did in FCP.


              I'm starting to think the issue has something to do with AE rendering to a file differently than the way it renders to the canvas. For example, here's a fast moving 50% opaque superwhite shape in the AE canvas:


              Screen shot 2010-10-19 at 8.23.55 AM.pngScreen shot 2010-10-19 at 8.45.14 AM.png

              I rendered this using Animation + Alpha using Best settings and 32-bit float, and reimported it into AE:

              Screen shot 2010-10-19 at 8.25.51 AM.png

              I think the answer may lie in the fact that the first image looks solid white in some places, even though it is 50% transparent. It's almost like the excessive RGB values "bleed" over to the alpha channel. The rendered image (probably correctly) looks grey because it represents white at 50% opacity.


              What I want to do though, is find a way to make my render look like my canvas. So it seems I need to do two things to the image:

              1. Find a way to increase opacity in areas of excessive white, so that they are not grey in the render.

              2. Find a way to "bake" the superwhite colors to a level below 100% white so that they show up in the render.


              I tried the "Alpha from Max Color" effect and this seems to get kind of close, but unfortunately it discards other parts of the alpha channel (like drop shadows).


              It might also be helpful to know that when I have AE output to an external NTSC monitor (via Intensity card) the monitor shows the exact same artifacts as the render.

              • 4. Re: How do you get 32-bit float effects to work with an alpha channel?
                Mylenium Most Valuable Participant

                There's nothing wrong with AE, you just have a wrong understanding of how float processing works. In short: You are mistaking blending for transparency, and that's why everything falls apart when you export. The color is still there, there is just no opaque pixels to process on and thus it never shows up, much less in an environment that only works in 8 and 16bit such as FCP, further limited by already killing fidelity by using an 8bit format. That's all that's happening and unfortunately there is really not much you can do. Unmultiplying the result based on luminance should give you a usable result, but it remains an approximation and no matter whether you use the same blending modes in FCP, it will never work the same. So in conclusion, a better way of doing this would be to work in 8 or 16bpc and make it look as if it were 32bpc. That is, if you cannot render the finals in AE, where again you would be able to get correct blending in float.



                • 5. Re: How do you get 32-bit float effects to work with an alpha channel?
                  Clint Porter Level 1

                  Thanks Mylenium for taking the time to write a detailed explanation. That's what I feared was the case. I did manage to get a fairly good result now by selectively applying the Alpha from Max Color effect in certain parts of the pipeline before darker elements are rendered. Maybe sometime I'll play with Pixel Bender coding and see if I can modify Alpha from Max Color so that it that adds to the alpha channel rather than replaces it.


                  Since almost all of my graphics go into my coworker's FCP timelines, I'll remember to only use 32-bit highlights in full screen graphics. Thanks again!

                  • 6. Re: How do you get 32-bit float effects to work with an alpha channel?
                    Navarro Parker Level 3

                    Just wanted to chime in with info that might not be obvious. When using the Animation codec, you are automatically bonking your final output to 8-bpc color. The only modern codecs to support higher color spaces are ProRes 422 and 4444. But even they are limited to 12-bit color. So ProRes can help keep smoother gradients and wide range of colors.


                    I don't know of any NLE than can natively edit 32-bpc video. At that point, you might at well edit everything in AE or Apple Motion.


                    One solution would be to edit in FCP, then use something like Boris XML Transfer to convert your FCP project into an AE project. Then do your final render in AE. I'm unsure of the FCP <-> Motion workflow.