Color management, workflow, and transforms

New user here.

I’ve made some video “slide shows” from photos using kdenlive, which look great. Now I’m adding video clips to the timelines, and I’m having some challenges with color management of the video clips.

I’ve just switched to shooting in Log (Panasonic VLog-L), which requires a transform to undo the applied log function. I’m using technical LUTs for this, since I don’t see another method in the manual.

I’m wondering how to sequence color correction modules - ideally, basic corrections should come before the technical LUT, and final touches applied after the LUT. But I do not know which of the color/image correction modules are suitable directly on Log format, and which should be reserved for after conversion.

So my initial color correction workflow looks like this:

  1. Exposure
  2. Color Temperature
  3. Lift/Gamma/Gain
  4. Saturation
  5. Color Channel Mixer
  6. HSL Primaries
  7. Apply LUT
  8. tbd…

Does this look reasonable? Also, does color correction in default REC709 maintain my 10-bit bit depth until the video is rendered?

Thanks in advance. I appreciate this great app and excellent documentation. :grinning_face:

1 Like

At present 10-bit and Log handling is still quite patchy, and I don’t think we have a clear and definitive list of what you can’t use if you don’t want to break that chain. It is on the roadmap to improve, so improvements will continue, but that involves lots of independent parties making improvements to lots of separate components.

I’m not sure that I would get too hung up on that though, mostly I would still just focus on doing what you need/want to do, and if the result looks ok, then it is OK. QED. If it isn’t, then you have something specific to investigate and report on. But you’ll always have your high depth masters, and your workflow, so you’ll also have the option of simply re-rendering that project with some later release that preserves that through more of the pipeline, if it really does make a notable difference, too.

So my initial color correction workflow looks like this

I don’t profess to be an expert colourist, but I do shoot a lot of underwater footage, and greatly prefer how things look in natural light, so I do spend a lot of time trying to make heavily colour biased footage shot in natural light look more like it was shot in White Light, and removing the contrast reducing effects of limited visibility.

So with that said, that effect stack looks a bit … busy … Do you really need and gain from all of those steps in what you are working on and trying to achieve? Some of them do much the same thing or are just heuristic shortcuts for settings in the more manual controls, so I’d be a bit worried you might be doing more ‘damage’ correcting the over correction of each previous step than you might be losing by dropping back to 8-bit colour somewhere in that line. That and it’s just so much more work to be juggling all of those individually than trying to do it with a smaller number of the more powerful controls (especially if you have them in the stack already).

So personally I’d be looking to simplify that stack. But colouring video is art, so really, do whatever it is that you think you need to do to get the effect you think you want in whatever way seems easiest or best to you. And don’t be afraid to experiment with different stacks of colour effects. I’ve evolved through several combinations as kdenlive gained new effects or or improvements to existing ones, each of which has its own advantages.

which modules should I put first and which afterward

There mostly are no hard and fast rules there either, but if you want to maximise quality then in general I’d be putting the effects which you are least likely to need to partially reverse in some later step first.

In retrospect I could probably use a curves (or Bézier curve) adjustment in place of the LUT, but my question remains the same - which modules should I put first and which afterward?

Thank you @Ron

I appreciate your reply. And I understand now that things are in flux re development.

I put in modules that I think I might need, but rarely use more than a few. Exposure, color temp, and saturation for example. But I like having options in place just in case. When I get a chance I’ll see if having extra modules causes performance to drag.

Thanks again for your thoughtful reply.

When filming with Log, it is usually a good idea to use a LUT. It is almost impossible to find the right settings manually (without LUTs). I also work with Panasonic, but I decided to use the CNED2 profile because it keeps the ISO number from getting too high and gives me the best tones and colors for portraits. I usually use the following tools:
White Balance (LMS space), White Balance, Levels, or better: Bézuier Curves, HSL Pirmaries, Satuation. But I also think it’s better to have less than too much.

Very good; thank you @micha.

I have the Pana LUT and a few alternatives. These are fine for a conversion from log-space to Rec-709.

But so far my colors are not good. For example I get strong color bands in the sky. Looks like poor 8-bit video or worse. If I leave out all adjustments other than LUT, video has low contrast and saturation.

Yes, I need to experiment more. Thanks for the input!

Experimenting is always good. For me too. Please write down which Lumix you use, which image style in the camera, and which LUT. Together, we can achieve better results.

Thank you!

I’m using the Lumix DC-G9, shooting in VLog L.

The Pana LUT is here: About LUT

There are other LUTs I have tried, but none solve my problem with contrast and “color breaking.”

I’ve tried some more. Starting with a clip that has the color-banding in the sky, I removed all color effects. This looks like log footage as expected. Adding the conversion LUT and Exposure after that results in color-banding. Bands trend from blue to magenta.

Next I grabbed a screenshot of the log footage using VLC. Format was 8-bit png. I imported this into my RAW photo editor (darktable) and tried some basic edits to the image. I wasn’t able to employ the cube LUT, so I did some simple things like boosting contrast, chroma, saturation, and applying a “curve” (tone equalizer) to brighten the highlights.

I can force banding if I try, but it’s just in the Luma channel - I do not get the color variation. And with a bit of restraint I can get a better-looking image. It’s not great, but I’m starting with a compressed 8-bit screenshot.

Back to kdenlive - I suspect that my main challenge is working with log footage in an 8-bit pipeline. Video shot in one of the standard formats looks a little better, but I lose some dynamic range (according to what I’ve read) and flexibility for color grading.

Hallo @Doug-Phoenix
It’s not that easy to produce better images with a log profile than with a well-chosen and well-exposed standard profile. Many people praise NAT, but it doesn’t work for me because it makes faces too red. PORT is even worse. I also opted for CNED or Cinelike D on my G9, which was the best the camera had to offer at the time. All my attempts with LUT have failed to improve the quality. However, with the so-called standard profiles, you have to expose well and sometimes also adjust the contrast when shooting. Then the results are the best; LUT is rarely really better. I gave up. With my S5M2, I only use CNED2 now. What I’m trying to say is that it’s difficult to process LUT well, and there may be situations where it’s the best option, but maybe you’re making it too difficult for yourself.

Hi @micha

I think you have the best answer. When I shot with Cinelike D things looked better. As you noted correct exposure is important, but that’s always true.

I appreciate your help.

You need to inform yourself: But I am convinced that with a normal image profile (NAT, CNED2), you can get a perfect shot if the contrast of the subject matches the profile and if it is correctly exposed. Log would not improve this shot at all. Log offers great advantages when you need more dynamics for high-contrast subjects or in other special cases. But the logarithmic curve has to be normalized again with special LUTs.

It wasn’t easy for me to give up Log when so many people talk about it so enthusiastically and my camera offers it too.

I think you are correct, if exposure is just right and you don’t need to push colors around very much. I think that is back to 8-bit color in the pipeline.

So this will be my approach with kdenlive.

OTOH, people do make a living doing things like shooting wedding videos, and pro’s often choose to shoot in vlog. So there must be good reason to go to the trouble.

Thanks again for your comments.

Hello @Doug-Phoenix
But you don’t necessarily have to film in 8-bit. Take a look at what your camera offers. I think the G9 can do it. I think: MOV 4K mit 10‑bit 4:2:2, 150 Mbit/s.

Even though Kdelive always drops to 8-bit as soon as you apply a color effect, it’s not that bad. I still always film in: MP4, Full-frame, 25p 420/10bit Long GOP, AAC 72Mbps. However, I always render in FHD 8-bit, h264, 25fps. If Kdenlive can handle 10-bit one day, I could use that.

Thanks again. Yes, I shoot in 4k 10-bit 4:2:2 at 150 Mbps as I recall. And rendering in 8-bit is ok.

I’m taking your advise above regarding recording color profile for use with kdenlive.

I do hope that one day full 10-bit color editing is integrated into kdenlive. It would be a major step forward for the project. But I appreciate that such things are challenging, and resources are limited.

1 Like