This got a lot more replies since I last posted so lets see if I can at least make a start with how I think a photographic scene referred workflow would look like. Since I do think the ACES standard[1] is probably a good reference to start from, lets first look at what ACES is and then try to compare the workflow to a photographic work flow.
The ACES standard is in its bare essence a workflow specification for managing color in a cinema production pipeline. To do this they introduced a set of color spaces
- ACES2065-1 - Main linear color space using AP0 primaries
- ACEScg - Linear color space using AP1 primaries (used for compositing and digital renders)
- ACEScc - Log based color space using AP1 primaries (used for color grading)
- ACEScct - Same as ACEScc but includes a toe in the transfer curve to make it more similar to older log based formate
Added to this are a set of well defined transfer functions for both input[2] and output, although for output there is an extra wrinkle in that instead of directly going trough output it first might go trough a Look transform and secondly trough what is called a “Reference Rendering Transform” (or RRT for short). The RRT is in practice directly combined with the output transform so isn’t directly seen.
On top of all this ACES provides a reference implementation of all this in an OpenColorIO (OCIO) configuration.
Chapter 1: Ingress
At this stage there is already a huge difference between photography and cinema, not in the first place because most digital cinema cameras don’t use RAW files but a high bit depth video file with a log encoded color space[3]. Also due to the almost ubiquitous use of stage light (even for outdoor shoots) with well defined properties a single common transform[4] can be used. It is at this stage as well that an initial look is decided upon which is then also immediately used for any monitoring and screenings. This look is later communicated to the editing/VFX/DI departments so that everyone agrees on what they are seeing.
Contrast this with photography where with a few exceptions (studio and product photography come to mind) there is a lot less control over the lightning conditions. Add to that, that the details of RAW files are often kept secret[5] and I know of no cameras that offer anything but 8-bit jpeg/tiff besides RAW. So before a photograph can be brought into a scene referred color space it first will need to be demosaiced, color corrected and transformed into the linear scene referred space (not necessarily in that order). Hopefully this can be mostly automated.
Chapter 2: Editing
In the cinema workflow the footage is send out to the different post processing departments that will use the look transform from the previous step in combination with a correct output transform to edit (select which footage to use in what order), add visual effects (compositing) and finally add the final color grade (the final look of the film).
In scene referred photography this wouldn’t be too different from the cinema space of course selection will look a bit different, so in practice only the color grade aspect would remain (unless doing a photo composite of course). In photography it will be generally be at this stage that a look with be developed.
Chapter 3: Egress
In cinema in this stage the output transform is applied for the specified outputs needed sometimes some extra grading is added here to adjust for some output specific hangups.
Except for the somewhat different output formats (film will hardly be printed on paper for example) there is not much of a difference here either for photography. The only problem might be that most printers only provide ICC profiles and not LUTs compatible with an OCIO configuration[6].
Chapter 4: What does this mean aka Conclusion
The biggest issue with scene referred editing in photography seems to me to be what I would like to call the “Ingress problem”, where a photographer will need a bit more control over the input transform then is currently given in a standard ACES workflow, even if this is mostly automated. Some other things that we as photographers might want to tinker with a bit more is the RRT (which is effectively the tone map operator).
In effect I would propose for a photography workflow instead of combining the RRT with an output to combine it with the look transform. Of course such a workflow would need to include a default look which could be based on the ACES RRT. In this case global adjustments should be done by changing this look transform to suit our needs on a photographic basis, of course local edits will also be needed but those will be done by adjusting the values directly in the scene referred color space.
[1] Surprisingly the last S in ACES doesn’t stand for Standard but for System
[2] Currently besides sRGB mostly defined for the log based formats many professional recorders/cameras work in
[3] This is effectively a sort of intermediate form between RAW and something like jpeg output, enough color and dynamic range data that further editing is possible without all that pesky demosaicing
[4] Although since each manufacturer has their own Log based color space you would need 1 transform per manufacturer in practice
[5] With the sole exception of cameras that can shoot DNG (like my own Pentax K-1)
[6] Although many modern printers want standard RGB files anyway and only provide the ICC file for soft proofing
?
)