RED footage and Exporting color accurate EXRs
Okay, here's the big picture. I'm overseeing a project that was shot with a RED WEAPON (image pipeline IPP2, RedWideGamutRGB, Log3G10). It's being edited in Premiere Pro. The typical pipeline at this facility is to roundtrip any VFX type shots (corner pinning, sky replacements, etc.) using AE and then send the locked picture via an XML to Resolve for the colorist who will then deliver baked ProRes files to be relinked in Premiere for conforming with the final audio mix.
So: Premiere>AE(some shots)>Premiere>Resolve>Premiere.
The issue is that the assigned colorist wants to keep everything as R3Ds for him to have maximum flexibility. He is rather set in his ways. He wants to grade only the original footage and then have the VFX team do their work after he's done with the grade.
So:Premiere>AE(some shots)>Premiere>Resolve>AE(some shots)>Premiere
That works okay for certain shots, but I'm sure you understand why that won't work as well for things like sky replacement because it essentially forces the compositor to take on colorist duties. It also means that in order to get to picture lock in the first place, the VFX shots have to be rendered (at least) twice: once before picture lock and then again after color.
I'd like to offer up an alternative approach to the colorist which would be to use After Effects for the VFX work upfront and render DPX or EXR sequences to be used in Resolve. I prefer EXRs, and so I've set up some tests but things are not working out as I had hoped.
Using Fusion, I loaded an R3D clip with the default RMD settings and rendered it out as a 16 bit EXR sequence. Loading the EXR sequence back into Fusion, I created a rectangular mask for the R3D "layer" and then overlaid it over the EXR "layer" using a merge node. The result: perfection. There is no visible difference between the R3D and EXR. Fusion is linear by default so this is what I expect given that both the R3D's and the resulting EXR formats are linear.
I bring both the R3D clip and the EXR sequence into After Effects.
Project Settings: Depth-32 bits per channel (float), Working Space-sRGB IED61966-2.1, √ Linearize Working Space.
RED interpret footage settings:
Assign Profile-Rec.709 Gamma 2.4
Interpret As Linear Light-On for 32bpc
EXR interpret footage settings:
Assign Profile-Rec.709 Gamma 2.4
Interpret As Linear Light-On for 32bpc
Same approach. I apply a rectangular mask to the R3D layered over the EXR. This time, the images are displayed differently. While similar, the EXR clip has brighter highlights and there is a visible color difference.
I've fiddled with settings every way I can think of but keep hitting a brick wall. I've tried other working spaces, including ProPhoto RGB to see if there might be a distortion or clipping of values, Preserve RGB checkbox on and off, etc. No dice. The problem seems to be how AE interprets the footage, or I just have the wrong settings applied.
Any insights? (Yes, I've read old posts but haven't come across this particular issue, though Color space and OPENexr Delivery advice needed by Christopher Stewart touches on the output side which I haven't even gotten to yet. I'd appreciate insight there as well if you have it to give.
Yes, doing VFX work on top of a final grade is a bad idea.
Have you tried setting the AE project Working Space to none, and 32bpc? Then when you bring in your EXRs and R3D, preserve their RGBs. You'll need to add a lin to log adjustment layer to the top of your layer stack, along with a LUT to view it properly. Just FYI, If you set your adjustment
layer to be a Guide layer, and you export EXRs, turn on preserve RGB to get proper half float EXRs for grading with. They will match your EXRs you imported exactly, except for the vfx work of course.
"Have you tried setting the AE project Working Space to none, and 32bpc? Then when you bring in your EXRs and R3D, preserve their RGBs."
I have. The issue persists.
Digging a bit deeper, it appears that it's the interpretation of the R3D files that is the issue. Here's the scopes in Fusion with the R3D on the left and the EXR on the right. (click to expand).
Note that they are the same.
Here are the same 2 files in AE with the AE project Working Space to none, and 32bpc. Preserve RGB is ticked for both.
First the EXR. Nice, as it should be.
Next the R3D. Something appears to be going on with the way the R3D is being interpreted such that the mids to the highlights are being compressed below 90 IRE.
I've also tried going into the custom R3D settings and making some adjustments, but 1) I shouldn't have to and 2) I can't access other parts of AE while I make changes so it's a bit of a fool's errand.
I've read previous replies from Andrew Somers to others and found them very enlightening but that knowledge hasn't helped me crack this puzzle.
Other thoughts out there?
have you tried creating the EXR with adobe only products? I feel fusion is the wildcard here. At least you'd be able to isolate the variable of EXR render output with assigned profiles.
I made an EXR from AE 32bpc and its luma matches, although the active window will be a little off so the pixels will be shifted a little bit. In Resolve, settings for levels should be set to auto so you don't export video levels(which might be your issue hiding in fusion)
There is also the free AE plugin opencolorio used by ACES for color management as well if you want to get complicated with ACES workflow in AE.
Stepping back is a good idea.
I've gone back to RED Cine-X and have exported EXR and DPX sequences with a variety of settings that seem like they would might work but I have yet to create files that are satisfactory in either Premiere or AE. I will keep working on it....
how are you comparing them? each program has its own particular flavor of color management. do they match if you import then into the same program window? are you using an external monitor? what is your monitor calibrated to?
I am a little unclear on what you need to be doing — did you just want to compare the EXRs you're generating out of Fusion to see that they match?
Based on the AE histogram, it looks like you have softclipping. But the upshot here is that you must have different settings for R3D import for AE vs what you are doing in Fusion.
Try: make a note of all the values for the R3D conversion in Fusion, and then apply those values in the R3D importer in AE (lower left "more" in the import dialog).
Adobe has a tendency to set their defaults to do things like soft clip, it seems clear that the RAW convert settings are different in the two apps.
"Preserve RGB" won't help because R3D are not RGB, they are RAW and need debayering first, and there are plenty of adjustments to control that, that can (and obviously are) different between AE and Fusion.
If your colorist wants "full control" then use EXR and not DPX. DPX is only 10 bit integer, EXR is 16 but float, and the difference is non trivial. USE EXR for intermedia steps whenever possible.
DPX is a depreciated format IMO, especially for intermediate stages.
Never use ProPhoto
ALSO: You mentioned once using ProPhoto - I recommend never, ever, using ProPhoto for any video/film project. The most important reason is that ProPhoto uses a D50 white point, and video/film is almost always D65 (D63 for DCI P3). Using a D50 whitepoint means that all your D65 footage has to undergo a Bradford matrix into D50, and will have to undergo that *again* from D50 to D65 on the way back out.
These needless transformations can result in a lot of unexpected color problems and artifacts. A second issue with ProPhoto is that is uses IMAGINARY primaries — the ProPhoto gamut includes colors that do not exist in reality including some that are NEGATIVE.
It is shocking to me that A Dough Bee suggests using ProPhoto in some of their documentation for color management and film. It should NEVER be used in our workflows. ProPhoto is mainly useful for photo editing for print destinations, as it allows you to work in RGB without clipping for a CMYK output.
FOR FILM: If you want to work in an ultra-big gamut, consider ACEScg, the cgi version of ACES, designed for VFX workflows.
VFX & Title Supervisor
Thanks all for the comments.
My goal is to be able to preserve the "look" of the R3Ds throughout the pipeline (Premiere, AE, and ideally Fusion, to Resolve) so that when the Colorist comes in the rendered VFX shots match the camera originals (as defined by the settings in the metadata.) Hence my desire to deliver EXRs.
I have figured out a somewhat workable process that works from AE to Resolve but I'm not satisfied with it. I will continue doing more testing and probably talk to Adobe, BM and RED about the various issues I've encountered. (Let the finger pointing begin! 😊) I hesitate to post it until I'm satisfied that it works the way I want it to.
I have considered using OCIO and bypassing AE color management entirely, but feel I shouldn't HAVE to. Perhaps that's where I'll end up but I'm going to tackle this puzzle a while longer.
[Jack Chavez] "R3Ds — In both AE and Fusion, I've started with the default interpretation, which references the metadata for deBayering, so to my mind they SHOULD match on the scopes, but don't for some reason. Perhaps AE is doing some soft-clipping.Since Fusion is a default linear workspace, there is no reason to default to a softclip. BUT
Since AE is not a default "linear" workspace, and many use it in a gamma encoded way, Adobe seems to keep their presets/defaults oriented toward that workflow. Linear in AE was something they added when they incorporated CMM/ICC, which....
[Jack Chavez] Didn't know that about ProPoto. I've seen it recommended in a few different places but, yeah, in the Adobe documentation for sure....Using ICC profiles and a color management engine seemed like a neat idea... at first.
Here's the problem: It's not well suited to video/film workflows.
Color Management using ICC profiles was created for the purpose of going from a D65 RGB color model to a D50 CMYK color model, using profiles specific to the output device.
To do that, the image data is transformed from D65 RGB into the D50 "Profile Connection Space" (PCS) which is either CIEXYZ or CIELAB. This requires both a Bradford transformation matrix (for white point), and the matrix into the PCS (though the math can be combined). Then it gets transformed to the CMYK space of the output device via another matrix/LUT of the profile. That's a lot of transformations, but was the solution to make a "one size mangles all" solution for desktop printing and prepress.
For video/film we don't need all that. We need a LUT to display on our computer monitor what the image will look like through the show LUT (the look that is developed for the show). OpenColorIO is one way to use LUTs instead of CMM/ICC profiles, and avoid relying on color management in AE.
The problem with ProPhoto, is that it was intended for photo editing for print, which is D50, and so by using ProPhoto you can work in RGB without clipping, and also not have to do the Bradford transformation when converting to CMYK etc.
ProPhoto is good for that, but bad for Video/Film (something along the lines of using a metric socket wrench to hammer rusty nails into a Ming vase)...
[Jack Chavez] If I export an EXR sequence out of AE with certain settings it does match the AE interpretation of the R3D. Same for Fusion to Fusion. Not for Fusion to AE so far. EXRs from RED Cine-X look nothing like the original image."This sounds like more or less expected behavior (not prefered behavior, just expected LOL).
My suggested workflow is:
Choose one method to create the EXRs, and create all that same way, then use the EXRs for the entire workflow.
VFX & Title Supervisor