Try the same test using Adobe Media Encoder.
Thanks, guys. Fuzzy: tried that before, along with pretty much every variation I could think of, and it doesn't help. Have you ever succesfully rendered out a DNxHD 444 quicktime in 'Trillions of Colors' using these options, though? If so, you're the first person I've found who managed to make it work, so it'd be great to know what version of AE you're using. Avid customer support have been seeing the same problem as I've had.
Todd: Media Encoder is fine at 32 bits but has the same problem at 48 bits. I'm afraid I'm a bit more camera-side than post-side, so I'm not totally sure how 32 and 48 bits equate to 10 bits - the obvious guess would be that setting AME at 32 bits is getting me a 10-bit DNxHD file, but I feel like it might be more complex than that. Is that something you can help with?
Sorry to take so long to reply!
I've done a quick test, and the summary information in the AE Project window isn't reliable: if you output a clip to DNxHD 8-bit, and select 'Millions of Colors' in the output module, it's still listed as having 'Trillions of Colors' when you bring it into After Effects... So unfortunately that summary info doesn't prove anything.
My thinking is that, unless you select 'Trillions of Colors' in the output module, you're not actually rendering out to 10-bit. And I haven't yet found anyone who's done that with DNxHD without getting a screwy output. Am I wrong?
Am I wrong?
Completely. Run the following test:
- set your AE project to 32 bit sRGB and linearise working space;
- transcode source footage to both DNxHD 422 and DNxHD 444;
- import transcoded footages into the project, drop them into the same composition above your source footage, set blending mode to 'Difference' and compare and contrast differences between DNxHD 422 vs Source footage and DNxHD 444 vs Source footage;
- transcode your original footage to any other 422 and 444 codec(s) you trust, import transcoded footages into the same project, drop into the same composition, set blending mode to 'Difference' and compare and contrast differences between Trusted 422 codec vs Source footage and DNxHD 422 vs Source footage as well as between Trusted 444 codec vs Source footage and DNxHD 444 vs Source footage.
Appreciate the speedy reply!
I'm not talking about the difference between 444 and 422, though, I'm talking about the difference between 'Trillions of colors' and 'Millions of colors' and whether that makes a difference to the bit depth within the same codec. If you export the same clip from After Effects to DNxHD 175 10-bit in two versions - one using 'Millions' and the other using 'Trillions' - there's definitely a difference, though it's too small to see with the naked eye using a 'difference' blend. And weirdly it's not as simple as one just having fewer colors...
- Fill the screen with a very subtle graduated fill which really tests the bit depth, e.g. 45% grey to 55% grey.
- Export that to DNxHD 10-bit using Millions and Trillions of colors.
- Bring them both back into AE.
- To really emphasise the subtle differences, apply a levels filter to both of them so the grad fill is stretched out to fill the entire range.
The 'Trillions' version has a smooth (though banded) gradient while the 'Millions' version has a lot of artifacts. Not exactly sure what's going on there, but the 'Millions' version is noticeably worse than the source footage. Which is why I want to get 'Trillions' working for DNxHD 444.
I'm not talking about something to see with naked eyes, I'm talking about technique, which allows you to find mathematical difference in pixels values, however small. And I offer you to compare that result with the result you can get out of your 'trusted 10-bit' codecs with 'Trillions' of colours. However, it's up to you whether to exploit it so as to test actual footage quality.
Here are the results of the test over radial ramp gradient in linearised 32-bit working space.
Difference between Ramp Gradient and DNxHD 422 8-bit 185:
Difference between Ramp Gradient and DNxHD 422 10-bit 185:
Difference between Ramp Gradient and DNxHD 444 10-bit:
Hey I'm not sure if this may still apply. I've been having the same problems. I used media encoder and set the "depth" setting to 32 bit. It seems to work just fine. Haven't tried it out in AE as yet.
Hope this helps.
K: there doesn't seem to be any difference at all between a 24-bit output and 32-bit output from Media Encoder. If you have a smooth gradient in your image, both seem to render it out in 4-colour-unit steps if you're using the 10-bit scale in the AE 'Info' window, if that makes any sense - i.e. if I run the cursor over my original greyscale gradient fill in AE it'll go from RGB 10-10-10 to 11-11-11 to 12-12-12, but if I run it over the version I rendered out through Media Encoder it goes from RGB 10-10-10 to 14-14-14 to 18-18-18. Which would make sense if it's actually only rendering in 8 bits.
Then if you set it 48-bits you get the funky image problems.
Fuzzy: thanks for the clarification - I see what you were saying, my mistake. But if you try comparing two outputs from After Effects, using exactly the same codec, but in one case selecting 'millions' and in the other selecting 'trillions' I think you'll see that you get very different results. Or at least I do. If you don't, then that's great, because it means it's just a problem with my setup.
I've spoken to Avid customer support, who spent a lot of time trying to work this out - they saw the same problems as I did and are mystified but don't have a solution right now. I also spoke to Adobe customer support, who - after a lot of back and forth - admitted that there might be some problems with 10-bit with DNxHD from AE and from Media Encoder, then said that DNxHD 'is not yet fully supported in this version', which was news to me... They've set up a support ticket for this problem, so who knows, they might come up with a solution. Fingers crossed.
But it looks like, as of this moment, you may not be able to export a true 10-bit DNxHD 444 quicktime file from CS6.
...then said that DNxHD 'is not yet fully supported in this version', which was news to me...
DNxHD in MXF container is not supported in CS6 and earlier.
...But if you try comparing two outputs from After Effects, using exactly the same codec, but in one case selecting 'millions' and in the other selecting 'trillions' I think you'll see that you get very different results...
I'm afraid I can't be more clearer: DO NOT interfere with Depth settings in both AE and AME.
Have you ever read the comments above?
Yes I have.
Yes. Thank you for being so kind.