Convert from ACES PSD to Adobe RGB PSD with layers and transparencies

​@cody_cuellar Thanks a lot for your detailed answer and your support.

Yes, both AdobeRGB (more precisely, eciRGB v2) and a TIF file with layers are requirements from our customer.

We render in Maya/V-Ray using ACEScg as render space. The multichannel 32bit EXR images contains mostly the beauty, an AO and matte passes. The matte passes allows us to edit e..g. only one specific material in the image. Normally, after editing, the image is converted using the "Duplicate To Profile" command and then reduced to 8 bits. The eciRGB v2 profile is then assigned.

However, a new customer would now like to have the reflection of the product in the floor as a separate layer. That was actually the reason for my question. I should have waited for my colleague; then he could have explained to me more quickly that the color space affects the blending of transparent pixels. And a blend looks different in ACES than in AdobeRGB.

Our current solution works very well and satisfies everyone. But we will definitely try your suggestion that we should adjust our OCIO config. I'm not a color space expert, so my question is: Can you recommend a linear AdobeRGB / eciRGB v2 color space?

Thanks for the detailed post — happy to help clarify this, but I have a couple of quick questions that will help get a better understanding of what you’re trying to do:

- Are you looking to do further compositing/editing in AdobeRGB for creative purposes, or is this more about delivering the layered file to a client or downstream application in that colorspace?
- Are the layers separate render passes (beauty, diffuse, specular, etc.), or are they composited layers with transparency that interact with each other?
- Is there a specific reason AdobeRGB is required — for example, a client spec or a particular application downstream?

The reason I ask is that converting a layered file with transparency between colorspaces isn't as straightforward as it might seem with multiple RGBA layers compositing together. When layers contain transparency, the way those pixels blend together depends on the colorspace the data lives in — linear ACES and gamma-encoded AdobeRGB handle that blending quite differently. So if we simply re-encode the RGB values into AdobeRGB, the edges and soft transparencies can look different even if the solid colors appear the same. That's also why “Duplicate To Profile” flattens the image — retaining layers while baking in a colorspace transform could cause the compositing to look very different.

The proper OCIO way to work in a target colorspace like AdobeRGB is to set it as your working space in the OCIO panel. When you do this, all smart object layers dynamically convert to that working space, and everything looks the same — with the caveat that transparency blending won't be fully correct, since gamma-encoded AdobeRGB handles alpha blending differently than linear. For more correct transparency blending, you'd want a linear AdobeRGB colorspace defined in your config instead. Either way, the key requirement is that your OCIO config actually defines the AdobeRGB colorspace you want to target — it won't appear as an option otherwise.

It's also worth noting that the profile you select in Duplicate To Profile currently isn't actually transforming the pixels into that colorspace — it's just attaching that profile as metadata to the output. It's on you to make sure the profile you choose matches the display you have selected, otherwise the result won't look as expected. This is something we're actively working on improving.

Once I know a bit more about what you're ultimately trying to achieve, I can walk you through the best approach for your situation if your current solution isn’t ideal.

​@Stephen Marsh Thank you again for your offer of help.

I see one thing I wrote was a bit misleading and unclear:

Tha actual values in the file are obviously recalculated in the conversion. That's not what I was referring to. The problem is numerical adjustments (manually set) and blend gamma (determined by the color space, or set manually).

But you got the gist of it anyway :-)

Thanks ​@D Fosse for your thoughts and hints. My colleague expressed something very similar today. I was likely somewhat blinded by the fact that when the image is flattened to a single layer, the results in ACES and Adobe RGB look almost identical.I overlooked the fact that things are actually a bit different with transparencies, since the pixels are calculated together and this depends on the color space.

In other words, we cannot simply "transfer" a 32-bit edit in the ACES color space—involving multiple layers—into the Adobe color space. Since our client requires an 8-bit image in the Adobe color space with layers, we must create the layers in the Adobe color space following the 32-bit editing.

That settles the matter. Thank you all for your help.

By the way: We work a Mac with PSD 27.7.0. 

​@Lars Gerstenmaier - Can you provide a download link to a layered file for testing?