Red Saturation Upon Export (Color Science Gen 5 & Rec..2020)

(edit)
SOLVED
TL;DR of the entire thread is: (1) don't use Rec.2020 if you aren't finishing in HDR (stick with Rec.709 2.4 for SDR); (2) Rec.2020's gamma is PQ not 2.4, stick with Gamma 2.4 for Rec.709; (3) don't use media players to test your exported footage and (4) calibrate your damn monitor!

--

Hi there,

I know this is a common issue for beginner/intermediate editors but I'm having an issue with a specific colour I've never had before: red saturation upon export.

I shot my footage in Blackmagic RAW on a BMPCC6K and I am using the "6K Film to Extended Video v4" LUT at the end.

To start this project, I decided to use Blackmagic Color Science Gen 5 and Rec.2020 from my usual Gen 4/Rec.709.

I export to ProRes HQ or Proxy as always and H.264 usually as well (although they show the same results).

I have screenshots of my panels, my nodes and the in Davinci versus export as well. I have no qualifiers or anything more complex than Camera Raw & Primaries (in two nodes) + the LUT.

Any insight into what may be going on? Give me any as technical as manuals or a simple as an option I missed, I have a computer science degree so I apologize if it's a simple as RTFM.

Thanks so much everyone!

Render-Issue-sml.jpg (1018.48 KiB) Viewed 2743 times

Link to PNG version on iCloud

[PS to admins: can we adjust the 1MB (JPG) & 2MB (PNG) upload limit for us video guys?]

LUT's are the worst possible way to convert from one color space to another.

My preference here would be to to work using ACES color management. Are you familiar at all?

Or RCM, for that matter. Both better than LUTs.

I thought you guys may be right with LUTs since I haven't been using them recently, but removing it and using custom contrast/pivot with a couple other adjustments lead me to the same general colour shift.

James-Alton-Visuals-2021-06-12-at-20.36.20-PM.jpg (980.47 KiB) Viewed 2450 times

Top left is with LUT, bottom left is my new edit without a LUT and top right is the preview window in Resolve.

I have a 27" 2019 iMac 5K/P3 display if that helps at all.

I just tried adjusting the 'Timeline color space' from "Rec.2020" to "Blackmagic Design Extended Video Gen 5" and saw no difference. I'll have to try ACES or RCM soon and report back the results.

You can't judge colors on a non-calibrated display in the viewer, not even on a Mac.

You can't judge colors on a non-calibrated display in the viewer, not even on a Mac.

But if the colour from the reference view is on my monitor and the export is on my monitor, why are they (that) different?

Its quite the shift from what I see to what is exported. Sure, my phone, iPad, other people's devices and the Youtube versions are all different, but they are all fed that red-shifted video, rather than what is in the reference window.

I assume what you're also saying is that I need to use something like an X-rite ColorChecker, ACES or RCS, trust my scopes and not my monitor... and that will help with finishing?

But there isn't much I can do right now with my setup?

You can't judge colors on a non-calibrated display in the viewer, not even on a Mac.

If your calibration removes relative differences, better get new device or seek help with calibration. Absolute accuracy won’t remove what op is seeing, it will probably magnify it even more.

@Hendrik
I don’t see that here on my setup.
@ the TO
How does a re-imported render compare?
BTW, such a redshift should be obvious on the scopes too.

Load that exported file back into Resolve and see if there is still a mismatch. Most of these mismatch problems stem from metadata interpretation in external video player, not from video data itself. To some extent it can be controlled by changing the qt atom values for exported video. There is the rec709-A setting for example, and options to set colorspace separately from color management settings in newer Resolve versions. But you can’t make it go away entirely because you have no control over those external players and implementations.

I am not a workflow/color space expert but something in your settings caught my attention.

Timeline color space: Rec.2020 Gamma 2.4
Deliver page color space and gamma tags "Same as Project"

AFAIK the vanilla standards are
SDR: Rec.709 Gamma 2.4
HDR: Rec.2020 PQ

So I don't know what to expect from rec.2020 gamma 2.4, where would you need that specific output?

Maybe the players you're using outside of Resolve are getting confused by this non-standard (someone correct me if I'm wrong?) combination of color space and gamma tags.

Since no monitor or screen can actually show the full gamut of Rec. 2020 it is normal to work in either rec. 709/BT1886 (100nits) for SDR and P3/D65/ST2084 (1000nits) for HDR as calibrated monitoring standards, furthermore your actually delivery. As to why your transforms are out, I would think it has something to do with your concatenation of luts - you are best to keep the translations as simple as possible Camera space/lut>workspace colour space> delivery same. I am finding it hard to follow exactly how you did it.

As has been mentioned, it is really essential to have calibrated monitoring. More important than using an Xrite colorchecker on the shoot is use a probe on whatever chain of monitoring you have, regardless if that is limited in capability. I would get an Xrite oem probe and use DisplayCal at the very least. Good luck.

But if the colour from the reference view is on my monitor and the export is on my monitor, why are they (that) different?

Media Players have their own potential to alter the image. That's why removing those variables is an important part of Quality Control.

The best path forward is using ACES or RCM paired with a Decklink and the highest quality calibrated display you can afford. That ensures image accuracy so your adjustments work.

Second best is ACES/RCM and taking the export from the computer to a calibrated display and play it from a hardware device. With this method, you'd take notes on image quality, make adjustments, and test another export. Less efficient, but...it does remove the variables that can alter the image.

The timeline colorspace should match the deliverable. Most of the color tools and transforms are color space aware so a 2020 timeline is going to be dangerous if not monitoring in that space.

Here is what's happening when using a 709 transform in 2020 space. Notice the vectorscope is nearly identical but the viewer (Use Mac display color profile) is reflecting the effects of the 2020 tag. See how the skin chips are all oversaturated yet it's not reflected in the vectorscope. Go back to 709 2.4 unless you are monitoring in HDR.

Good Luck

7092020.jpg (109.3 KiB) Viewed 2330 times

Here is the same footage correctly mapped into 2020. The viewer is correct but now the vector, which I believe is always 709 scene is wonky. The CIE scope correctly reflects timeline changes but it's more neat that truly useful.

Good Luck

2020.png (231.6 KiB) Viewed 2321 times

(I just wanted to start off by saying thank you to everyone who has taken the time to reply. Clearly I need a little upgrade in professional tools and in my professional training, so I appreciate you meeting me where I am now and helping me out.)

To answer the first comments about scopes and re-importing the footage (@Hendrik Proosa & @Uli Plank
): I previously white balanced the footage using the scopes and the oddly enough, re-imported footage looks the exact same (no colour shift, same scoped).

The colour shift could be then, as you are saying, Quicktime's fault as it could interpret the colour differently than Davinci intended. I tried VLC and took a screenshot but it is even more red-shifted than Quicktime.

About my mismatching settings (@ttakala): Thank you for noticing that! In my amateur research I figured shifting from Rec.709 to Rec.2020 would help preserve more colours in my colour space, but clearly I didn't know how to do that. Like an amateur, I just changed one setting without looking at others.

So I adjusted my timeline colour space to ACES and my output to Rec.709 Gamma 2.4 and this, for the most part, fixed the issue!

EDIT: It looks like iCloud is a terrible image host so the "solution" (Rec.709 Gamma 2.4) is here for easy viewing,

James-Alton-Visuals-2021-06-13-at-09.37.00-AM.jpg (782.99 KiB) Viewed 2310 times

It still has some colour changes (now by reducing red saturation and adding contrast) but that is more expected in a conversion from ACES to Rec.709 than what was happening before. I re-imported the footage into Davinci and the scopes were the exact same, so this could be Quicktime producing the shift again.

Thank you for listing the standards @ttkala. It's funny, I read how Rec.2020 has a colour space of PQ and not gamma 2.4, but it never dawned on my how to adjust the output accordingly. I think in general, I am unaware of how to utilize Rec.2020 with my BMPCC6K footage and since I am not finishing for HDR, I should stick with Rec.709 Gamma 2.4 on the timeline and for my finish.

@Steve Fishwick, AFAIK, I imported the footage into my timeline, made node adjustments to [Node 1:] Highlights/Shadows, [Node 2:] Saturation/Tint/Temp, [Node 3:] qualifier for face to re-add warmth after WB, [Node 4:] Contrast/Pivot, [Node 5:] Mid Details. Then export. Not sure if there is an issue with that but you were definitely correct on each player (Quicktime, VLC, etc.) having its own colour space interpretation and not a good representation of colour space. I am on picking up a Xrite oem probe now, thank you so much for the suggestion!

@Jim Simon, you are absolutely correct about media players and colour shifting. I didn't even realize this as a possibility (duh) and it will no longer be a part of my export process. I am working on calibrating my display and will look a DeckLink too for playback. I really appreciate the help here.

@Howard Roll, "Go back to 709 2.4 unless you are monitoring in HDR." This sums up the entire thread and exactly what has fixed the issue. Colourshift in the skintones. I am only finishing in SDR so there is no need for Rec.2020 in my workflow. I thought mathematically, "more colour in my colourspace, the better the output," but clearly with compression, shifting and interpretation changes, this is not true. Edit for what you're finishing in. Switching to Rec.709 2.4 in my output fixed the issue, now I'll change it for my timeline going forward as well. Also what is that colour chart you are using?

The TL;DR of the entire thread is: don't use Rec.2020 if you aren't finishing in HDR (stick with Rec.709 2.4 for SDR), Rec.2020's gamma is PQ not 2.4, don't use media players to test your exported footage and calibrate your damn monitor.

Thank you so much everyone! I really appreciate all the help. I thought I was an intermediate editor but clearly I need more education, practice and adjust my workflow with some more professional tools.

Also what is that colour chart you are using?

Its a Chroma Du Monde 12+4R, they're nice for checking cameras and sanity.

Rec.2020's gamma is PQ not 2.4

Originally 2020 used the same transfer funtion as 709. Rec2020 can also use a 2.4 (suggested SDR monitoring gamma), PQ, or HLG. The newer Rec2100 eschews SDR transfer functions in favor of PQ and HLG.

Good Luck