Resolve interpreting alpha channel wrongly

Hello,

I've been through a hilarious situation regarding the interpretation of alphas and mattes that Davinci Resolve does.

I have an .exr GFX shot (graphics over black) with an alpha channel. My project is set in ACES1.3 and the shot was render in ACES 2065- AP0.

So when I import the file and drag it into the timeline, Resolve identifies its alpha channel but the level of transparency is not the same as the one setted up in Nuke (the grey parts of the alpha mainly) . I have tried the premultiplied and straight options for the alpha.

The hilarious part comes in when I try the same GFX in Fusion. Just with a merge node the result is identical to the one designed in Nuke.

If this was a personal project I wouldn't mind to go through Fusion. But this is a delivery to a client that wants their GFX to be set in place in the timeline just by dragging the clip.

*Extra bullet: Looking for a workaround I tried to separate the Alpha channel as a matte (two files) and set everything up in the Color page (GFX plate + external matte). The result was the same that the one I got in the Edit page.


Do you guys have any ideas of what could be happening?

I would extremely appreciate if anyone could help me with this.

Thanks in advance.
Best,
Gabo.

Are you working in a color managed environment? If so, is it possible the color space / gamma of the exr is being mis-identified or somehow tone-mapped? I'm waiting for Shebbe to respond to this so that I can learn something, lol.

I believe Resolve Automatically interprets EXRs in an ACES project as ACEScg. Also, it's possible if your project is set to ACES AP1, and not AP0 and it's interpreting your EXRs incorrectly.

Sorry it's not an answer, as I'm not sure. I'm just trying to go through the things I would check (which is sounds like you've tried a lot already).

Are you working in a color managed environment? If so, is it possible the color space / gamma of the exr is being mis-identified or somehow tone-mapped? I'm waiting for Shebbe to respond to this so that I can learn something, lol.

Hello Steve,

I'm working on a ACES1.3 Color Managed project and indeed it seems like the Edit and Color pages are mis-identifying the gamma of the alpha channel.

I believe Resolve Automatically interprets EXRs in an ACES project as ACEScg. Also, it's possible if your project is set to ACES AP1, and not AP0 and it's interpreting your EXRs incorrectly.

Sorry it's not an answer, as I'm not sure. I'm just trying to go through the things I would check (which is sounds like you've tried a lot already).

Hello Ryan,

The project is set in AP0, in any case switching to AP1 doesn't seem to be affecting the color/alpha.
Following what you mentioned I've tried to apply and a different ACES IDT to the .exr resulting no solution.

¡Thanks to both of you!

I'll keep looking for a solution and of course sharing in this topic any discovery made.

Let's see if anyone else has an idea on what's going on here.

Cheers,

Gabo.

Two possible causes, can be happeming at once too: merging happens in nonlinear image data state and/or alpha channel goes through colorspace transform (which is obviously a bug).

Are you simply loading the exr in Fusion as-is or do you add colorspace conversions? And do you compare it to Nuke in OCIO ACES project or plain Nuke default? It won't fix your problem but gives a baseline in Fusion to validate edit page against. You need proper ACES CM in Nuke to validate ACES based workflows.

Two possible causes, can be happeming at once too: merging happens in nonlinear image data state and/or alpha channel goes through colorspace transform (which is obviously a bug).

Are you simply loading the exr in Fusion as-is or do you add colorspace conversions? And do you compare it to Nuke in OCIO ACES project or plain Nuke default? It won't fix your problem but gives a baseline in Fusion to validate edit page against. You need proper ACES CM in Nuke to validate ACES based workflows.

Hello Hendrik,

Thank you for your reply. I'm loading the .exr as it is without any conversion, just the media node straight into the merge node, the result as I've mentioned was exactly the same I got in Nuke. The Nuke project I'm compraring to is set in OCIO ACES, so I'm assuming there is something wrong with the interpretation that the Resolve's Edit page does.

Hello everyone,


I've just noticed that this problem only emerges when using an ACES Color Managed Project. By switching the project to REC709 the problem regarding the transparency disappears completely, you would get the same result in Nuke/Resolve/Fusion/AE and so on.

Does anyone knows if there is any ACES workaround that I might be missing to interpretate my mattes correctly?