I’m quite new in ACES world and I have a lot to learn for sure. But I have already had that awkward feeling when working for an sRGB ODT. Somehow it feels sometimes I’m fighting, inside Davinci Resolve, against the harder contrast the RRT+ODT impose, specially in dark scenes…
Things get clearer when you see what’s going on on a grey ramp’s waveform.
I guess we should get used to it and learn how to deal with it, shot by shot. But meanwhile I found this workaround:
I use two nodes with AcesTransform OFX, the first converting ACEScct to Rec709, and the second one converting sRGB to ACEScct. I’ve found this way I can rectify the curve and also get the possibility of putting a node between them to make some primary grading in davinci’s normal behavior (although I don’t know if that is such a good idea…).
Wether this method messes the image up in any destructive way or not, I cannot tell yet. So far is working fine. But you tell me if I’m doing something wrong and I’m not seeing it.
I think what you are referring to as “Davinci’s normal behavior” is just output-referred grading. As in - you are manipulating the same code values that are being sent to the display.
When using ACES, you are grading scene-referred and viewing the effect of those changes in a display-referred space (actually you’re grading on a logarithmic representation of the scene-referred data so that the film-like grading controls respond familiarly). Some people call this grading “under a LUT”. It’s akin to many other common non-video workflows.
If you’re used to display-referred grading, ACES (and other non-display-referred color pipelines) can feel a bit weird because the waveform is showing you the output and not the values that you are affecting (in display-referred these are the same thing and so the controls feel more directly “attached” to what you’re seeing).
If your anticipated deliverables are only for one specific target display, then display-referred might make sense for you. However, if you ever plan to deliver output for any other display, grading scene-referred using ACES makes a lot more sense as I believe it facilitates making multiple deliverables more easily.
for sure it can lead to problems with colors that are initially out of gamut from Rec.709 and you will certainly loose some image quality with your described workaround.
You basically work around the ACES pipeline.
I put together a little article what it means to move from DaVinci-YRGB to ACES 1.1 in Resolve.
Thank you Scott for the info, very useful indeed.
Yes. I think I should have put “normal behavior” between quotations marks…
As you have said, used as I’m to display-referred grading, it feels very weird to manipulate davinci’s controls now… not only when watching my waveform, but also on what the image results concerns.
In fact lift/gamma/gain controls seem to become useless tools now…
I’m trying log controls, which seem to work better, although I’ve heard they can be quite destructive if they’re no used carefully. Any thoughts about that?
A second question: what do you mean when you say “film-like grading controls”?
And still another one: regardless davinci’s controls “normal behavior”, what do you think of the pipeline (cct>709 >>> sRGB>cct) I’m using to rectify the sRGB curve? Could it be a problem? Does it worth it?
Thanks again!!!
Ignacio
Hi Daniel.
Thanks for your reply.
I had already read your article a few weeks ago and found it illuminating.
I have indeed had some particular problems grading between the color space conversion nodes, specially in secondary corrections (torn apart gradients in color range selections, and that sort of things). But I haven’t seen yet anything weird working outside those two nodes, and it feels quite confortable to have the sRGB-ODT curve rectified that way. Perhaps I’m missing something?
Thanks again!
Ignacio
The process chain in ACES should be always:
whatever–>ACES (AP0)
work in the working space ACEScg/ACEScct (AP1)
and only at the end of the process compress and clamp value for the output device (ODT).
Any 3D-LUT or color space transform that converts and reduces the image data for a “display” in the middle of the process, therefore must be lossy and should be avoided.
If you try to think working in the DaVinci in the default color management, but "unde"r a 3D-Lut for an Alexa for example in the first serial node, then you work similar like ACES, before the LogCtoRec709 LUT.
But here you can work also after the LUT if you want to, but the dynamic range of the camera material is lost.
The same happens when you introduce such a workflow into ACES. Only that in ACES the ODT is and should be always the last operation going from scene referred to display referred to avoid any loss of data.
Think about the ODT as your sound mixdown for a small room or a big stage. You don’t want to use any of that mixes to do another remix - you want to use the source files of course.
I am not a colorist but I know about the struggle of colorist grading in ACES. I think its more a problem of how the tools are used in Resolve, they are still based on old film to video telecine workflows.
As far as I know Baselight went another route with BaseGrade.
Thanks again Daniel for such a complete explanation!
I knew, of course, 3D LUTS could be destructive, because of the the limited information they use and the consequent interpolation process they make. But I thought Color Space Transformations were not destructive, since they are (or so I thought) mathematical operations performed on image codevalues.
I suppose not everything said in youtube tutorials is true…
I was quite excited about my pipeline. I thought avoiding corrections between the two Color Space Transformation nodes will keep me safe from any lost of information, with the advantage of getting a straight output curve for sRGB ODT.
Do you know of any other way for rectifying that curve?
Thanks a lot!!!
Ignacio
I think the question is, why do you want to change the behavior in the first place?
I am not a colorist, I started on flame and later continued working with Nuke&Flame.
You can work scene referred in flame, but most people I know work display referred up to today.
Whereas in Nuke everyone worked scene referred since the beginning, but I guess same as me, didn’t even understood it for a while
The two approaches are quite different, but after a while you get used to them. And I highly prefer to work scene referred over display referred if I am able to and it makes sense for the task.
I assume with Resolve the issue is similar when the Resolve color management and ACES got introduced in the software.
The moment you work in Resolve DaVinci-YRGB and apply a 3D-LUT for ALEXA footage LogC2Rec709 and grade before the LUT, the controls in Resolve should behave quite similar like working in ACES.
One big difference that I am aware of is, that in Nuke I am always working with material which is linearized, whereas in Resolve the material is presented to the user controls via a log curve. And in Nuke the view transform is always at the end of the process whereas in Resolve depending on the color management settings you are able to work after the view transform if you want to.
What is this plot showing exactly? I’m not exactly clear on what issue you’re trying to address. Is this about the grading behavior or something else?
This seems like a weird chain of transforms. Again, I’m not sure exactly what you are trying to rectify. Is it just that the sRGB curve is too contrasty for your liking? Is it the grading behavior? Is it something else? And you grading in between the 709 and sRGB transforms depicted in your processing chain?
Well… Davinci’s grading behavior in ACES was my first concern, as I found some tools I usually used (lift/gamma/gain bars) weren’t useful anymore. But digging deeper in the subject I learned that’s something I could cope with (or I guess so…).
But I still find puzzling to begin grading from a curve (or “for” a curve, I still don’t understand that) that’s already bent down. In dark scenes it feels (and may be it’s not more than that, a subjective feeling) I have to struggle to get what I want in shadows…
Thanks again Daniel:
I think you’ve explained it very well indeed.
At this point I understand I’ll have to get used to how davinci behaves when grading scene-referred, and I’m already working on that: I’m right now trying log wheels, testing low and hi ranges, contrast values, pivot point, etc., and beginning to get more of it (or I think so), as I’m also getting used to work with that sRGB output.
I have just read your articles which were both very useful to understand a bit more of the whole thing.
Thanks a lot for your help!
Kindly
Ignacio
