Confusion on colour grading and a workflow (with AgX + DaVinci Resolve)

[Velepexon]

By the way, I need to mention this: I am working with ripped game assets, so I’m limiting some details here. Fortunately, my problem is (hopefully) only about grading, so this shouldn’t be an issue outside of providing examples of my scenes. Any renders I’m uploading would have the ripped game assets completely hidden, which will lead to some empty spaces. I hope that doesn’t become a problem.

---

Hi, I’ve been unsure on where to ask help for this, thought I’d try here. Since around late August last year, I’ve had the motivation to try and learn more about colour management and colour grading. I felt it wasn’t enough to just export a final render in Base Filmic or (pre-4.0) Base AgX. I have accidentally fell into a few rabbit holes, mainly ACES, and while I would love to find more clarification on those sort of things, I believe I should be prioritising what I want to use anyway.

So, what I have at the moment is Blender 4.3 and the free version of DaVinci Resolve. I’ve been playing around and finding tutorials online, but now I’m stuck at a point where I’m not so sure what to do with the information I’ve gathered, nor can I confirm if the information is even accurate.

What my current setup is: I render the scene as a Linear Rec.709 EXR (Half Float, DWAA Compression). I import it into DaVinci Resolve where I then use an OCIO Config node in the Fusion page to convert from Linear Rec.709 to AgX Log. Then, in the Colour page, I do the usual grading (Primaries, Contrast, Saturation). I usually apply a basic Glow after these changes. Afterwards, I use a baked LUT that converts AgX Log back to AgX Base sRGB, giving me a finalised image.

Base AgX (what I see when it was rendered)

After Grading

Now, this workflow has worked pretty well for me. I get results that, to my eyes, look pretty nice. But something about it just feels suspicious to me. This is where it’s going to be tricky, because I don’t know how much of this stuff is true, how much is nonsense, or just what am I misunderstanding.

I’ve heard colour adjustments, especially on a perceptual level, should be done in log, but in other scenarios, the colour adjustments would be done after converting to AgX Base. So, is it one or the other, or can it be both?

I hear the main reason of using Linear is when needing to work with “absolute values,” like with exposure. But I’m doing my Primaries/Exposure in Log, so is that wrong, then? Can I be sure that if I set the “Gamma” of the node in Resolve to “Linear,” it is working as expected? Or should I be doing Exposure adjustments in Fusion, where I still have the Linear values from the EXR, before I convert to AgX Log? Should I even be using AgX Log for grading, or should I be converting to AgX Base?

Same goes for adding Glow/Bloom, that sounds like it should be added using the Linear values, instead of after converting to AgX Base. But I’m not entirely sure where I could even put it in Resolve. Should I add Glare in Blender’s compositor before exporting as an EXR?

This is all the questions and confusion that I can think of at the moment, and I’m sorry if this isn’t entirely the right place to have all of these questions, and I’m sorry if it’s not the best organised… Just this morning I only felt I should at least try asking somewhere. I’d appreciate any help, and I’m sorry if some stuff could’ve been researched better…

[oo_1942 (Kim)]

If I save it as AGX and convert DaVinciResolve back to AGX, wouldn’t I be doing AGX conversion twice?

I think I need to save it as sRGB Standard in blender and then switch to AGX in DaVinciResolve to see the normal color.

[Velepexon]

Saving the render as an EXR file should automatically save it as Linear Rec.709, or in other words “Raw.” So, there is no double AgX conversion in DaVinci.

[oo_1942 (Kim)]

EXR is stored as Standard without being affected by Color Management.
I didn’t use EXR much, so I didn’t know.

I also newly learned that converting from Composing is necessary for EXR storage from blender to AGX.

Add…

However, EXR is saved as Standard, not Raw.
Raw can be saved using the Tif format.
I don’t know if there’s any other way.

[Velepexon]

EXR is saved as Raw, you just have Standard in your View Transform.
Did you check to play around with more View Transforms?

Edits:

Edit 2: There IS an option to Override the colour management of an EXR file, but I don’t have any clue what use would there be for the Override other than maybe if you wanna export it as ACEScg, maybe.

[RobWu (Rob)]

[Velepexon]

Interpretation of answers based from just the teaser alone.
Edit 2: Added additional info from the course RobWu linked.

I’ve heard colour adjustments, especially on a perceptual level, should be done in log, but in other scenarios, the colour adjustments would be done after converting to AgX Base. So, is it one or the other, or can it be both?

It can be both.

Or should I be doing Exposure adjustments in Fusion, where I still have the Linear values from the EXR, before I convert to AgX Log?

If I don’t figure out how to convert Linear to Log in the Colour page without clipping values (or the solution requires the paid version of Resolve, which I’m sure the course will mention DCTL’s), then this can be a workable alternative.

Should I even be using AgX Log for grading, or should I be converting to AgX Base?

For the first half of the question, the course uses either ARRI Log C4 AWG4 or DaVinci Wide Gamut Intermediate. Since with my past tests, AgX Log “looks” close enough to DaVinci Wide Gamut, I will interpret that I can still use AgX Log for grading. The second half of the question, the answer is still yes since I want to keep using AgX. And it connects with the first question as well, I should be allowed be to make edits in the Linear “space” (cough cough), the Log “space” (cough cough cough) and then in AgX. Should be pretty simple.

Same goes for adding Glow/Bloom, that sounds like it should be added using the Linear values, instead of after converting to AgX Base. But I’m not entirely sure where I could even put it in Resolve. Should I add Glare in Blender’s compositor before exporting as an EXR?

Glow/Glare/Bloom happens in Linear “space” (excuse me) before converting to Log or Base AgX.

---

Appreciate it!

---

Edit 1: Skimmed through all 4 hours of the course, I recognise 96% of what’s going on here. I’ll try out some of the stuff mentioned here and experiment with them. I’ll mark your reply as a solution since it technically answered some of the questions I had. Still, though, thanks for sharing it.

[oo_1942 (Kim)]

I use more color space and PNG low, so it’s not a big problem

[SteffenD (Steffen Dünner)]

BTW, if you want to apply photometric operations in DaVinci like e.g. change the exposure down by 1 (= half the light reaches the sensor so to say) this of course should happen in a linear gamma to match what e.g. a DOP would expect.

This can either be done

• by setting the grade node’s Gamma to “linear” and then use the normal gain wheel set it to 0.5 to linearly half the amount of light
• or by leaving everything as is in your working / timeline space and using the Global wheel in the HDR palette and set the Exposure to -1 (just like in Blender’s display color management).

The result is the same.

[Velepexon]

• by setting the grade node’s Gamma to “linear” and then use the normal gain wheel set it to 0.5 to linearly half the amount of light

This was one of the questions I had in the post. So, is it that regardless of what is being inputted into the node, as long as the Gamma is set to “Linear,” the Gain wheel, as explained, should work as expected? Because where I’m getting confused is I’m still messing with the image while it’s in a a log state, and it seemed weird to me how I’m still within log but I’m adjusting linearly, if you get what I mean.

[SteffenD (Steffen Dünner)]

Some of the tools in DaVinci are color space aware if you use Resolve Color Management. Some are even if you use a CST workflow like I do (because I want to know what exactly goes in and what goes out without a “helping hand”. I’m used to this workflow from Fusion where you have to do everything yourself, even linearizing 8-bit LDR footage unlike in Nuke ).
But yes, if you set the gamma of a node to linear, everything inside is treated in linear math. Some controls do react rather funky when used in non-log space.

In the end I would prefer the HDR palette for photometric actions or something like white balance while the log wheels are more suited for artistic / lookdev tasks. YMMV though.

I can only highly recommend Cullen Kelly’s Youtube channel. I learned about grading linear there:

When you should color grade in log vs. linear | DaVinci Resolve

I stole this trick from VFX artists

And don’t get fooled by the sometimes quite clickbaity titles of his videos. He knows what he’s talking about.

[Velepexon]

Yes! I love Cullen’s videos a lot. I should really go back and rewatch some of his stuff, I might be getting a bit rusty. But thanks for the clarification!

[SteffenD (Steffen Dünner)]

I just downloaded this very interesting course, watched part 1 (“EXR Blender to Resolve Basics”) and noticed some stuff that I don’t find helpful or rather misleading IMHO.

For example: They show how to set up Resolve to preview linear EXRs from a renderer to give similar results like Filmic view transform in Blender by importing a fire sequence and using a single CST (color space transform node).
Filmic indeed isn’t the best possible way and gives this yellowish result on the right compared to even worse single node CST on the left:

Later they show the result with AgX, that looks completely different. The fire is more red (which is much more correct indeed) instead of yellow:

You could try fiddle with any knob on the single CST and never achieve anything like this in Resolve. So they take the deep dive and continue to use ARRI Wide Gamut plus an AgX LUT in the end.

BUT: If you use a proper 2-node CST workflow, you instantly get very close results even without having to deal with external LUTs or using the Blender ocio.config in Resolve’s Fusion page.

Just use 2 serial CSTs like this:

• The first CST to convert from Rec.709 linear to DaVinci Wide Gamut / DaVinci Intermediate like this. Notice that nothing gets “tonemapped” yet. The whole data is just dumped into a giant color space / log gamma:

• The second CST to convert and tonemap from DWG to Rec.709 / sRGB gamma:

Here the full range of the DaVince Intermediate log (10000nits) is tone mapped to Rec.709 with sRGB gamma using “Luminance Mapping” which gives a very close result compared to plain AgX:

You could also use “DaVinci” tone mapping that gives a slightly more orange result.

Long story short: In the end it’s a second, super simple CST to give much better results. BTW you could also perfectly adjust exposure with a color correction node between the 2 CSTs when either setting this color correction node to linear gamma or by just using the HDR sliders instead of the usual log ones (like I posted in this thread before).

Then they show the mushroom example and say: If you set the CST to sRGB you get results comparable to Filmic, while when setting the gamma to 2.4 you get more like an AgX result. This is complete nonsense IMHO.

The course itself is absolutely brilliant though and will help many people with finally ditching rendering PNGs and then trying to “color grade” them

P.S.: I used Fusion to take the screenshots, but the CST nodes have exactly the same controls in Resolve and they also look the same. Just in case someone is wondering.

[Velepexon]

Can’t go 5 minutes without colour management issues, hahaha

Very interesting addition here, I actually would’ve missed this entirely. I did felt it was weird how AgX had such a red look to it compared to the other examples, but shrugged it off since I was skimming a lot.

The colours do be colouring

[RobWu (Rob)]

To be honest, I’ve only seen the ‘ads’ for this one, and haven’t watched any of it yet.
But having Gleb saying on video that they had extensive talks with people like Troy about all this, I wasn’t expecting your comments on this
Will have to dive into it myself soon I guess to see.

edit:
I also feel that the way Blender works with the View Transform and Look can be very confusing for newcomers in 3D, and coming in from other 3D apps.
I rather have a ‘clean’ Rec709 render, and have all the other stuff in the compositor to fiddle with.

[SteffenD (Steffen Dünner)]

After watching the full course, I have some hopefully helpful hints for anyone using Resolve, that weren’t mentioned by Gleb during the course.

• When using the LUTs, you should set their interpolation to “Tetrahedral” instead of “Trilinear” to get a better interpolation / less artefacts:
  
• You can simply double-click any value name in Resolve (and in Fusion as well) to reset it back to default. No need to click into the value field and manually type in “0” or whatever there. This also goes for values that don’t have the explicit “Reset” logo () next to them.
  And it also works with color sliders and wheels.
• When talking about the light layers (video #04 of the course) it looks like he’s manually rendering out “OpenEXR” (not “OpenEXR Multilayer”) sequences for each light. I wonder why he isn’t using OpenEXR multilayer with all the light groups being spit out simultaneously in one go. Resolve can access these layers via the Fusion page very easily. It might be possible without the Fusion page as well, but that’s how I got it working:
  
  This also has the advantage that you need to only replace a single source when a new version of the rendering is done.
• He stacks the clips on the edit page and sets their “Composite Mode” from “Normal” to “Add” one clip after the other. You could also just select all the clips at once and change their composite modes on one to have them set for all (you don’t even have to press ALT like in Blender ).

As I said, the course itself is awesome and will bring the unavoidable world of color management a lot closer to us Vfx guys. Great work!

[SteffenD (Steffen Dünner)]

@RobWu

But having Gleb saying on video that they had extensive talks with people like Troy about all this, I wasn’t expecting your comments on this

I had the same feeling before I started watching it. The name “Troy Sobotka” alone should make all that he says bulletproof and true
The course itself is great, I just noticed a few things that kind of slipped through the QC process so to say and that could make the complex topic of color spaces, color management etc. unnecessarily harder than it needs to be IMHO.
For a potentially free (or “freemium” as they call it) course it’s the best you can get for sure.