Blender AgX in darktable (proof of concept)

Christian_Pfister · November 8, 2025, 3:48pm

A bit off topic, but I use the saturation slider in AgX only from time to time and only slightly (110-115%). Setting contrast well in AgX helps already to give a good push at colors. I use now color lookup table more often or the primaries in AgX to tweak specific colors. I also use color balance rgb, but now mainly for local work with masks.

Jetze · November 8, 2025, 4:20pm

Same here, prefer AGX now as it is fast and flexible. Still have to learn though.
Switched off my automatically applied preset of ColBalRGB.
Kind regards, Jetze

MStraeten · November 8, 2025, 7:15pm

Also use colorbalanceRGB. AgX is a tone mapper, not the module to rule them all.
You might use the saturation slider in AgX but that’s in the display referred part and so can introduce issues

Jetze · November 8, 2025, 7:22pm

Thanks Martin, guess you’re very right to say this.

Will not forget colorbalance rgb as I greatly like the module. But during the first part of the learning phase of AgX I won’t make much use of it. For me dt has quite a learning curve - which by the way I do love.

Kind regards, Jetze

Masterpiga · November 8, 2025, 7:36pm

Indeed, I end up not using the saturation in agx because in general there is very little margin for maneuver before artifacts start creeping in.

But contrast, slope and lift are my best friends

Jetze · November 8, 2025, 7:50pm

Ha, enlighten me please about the artifacts so that I’ll know what to look for.

Masterpiga · November 8, 2025, 8:20pm

Well, some kind of posterization which I don’t see if I increase the saturation in color balance rgb. I don’t have much more to add because I tried a couple of time and I didn’t like the result, so I just stopped using it.

Jetze · November 8, 2025, 8:36pm

Thanks, I know now what to look for and will inspect better my recent results!

stuntflyer · November 8, 2025, 8:38pm

@MStraeten, It would be helpful for me and perhaps others to know which AgX sliders are display referred and which scene referred? Does this suggest that as a tone mapper, AgX should not necessarily used for doing everything in this one module. I have read where some other people are doing just that.

mino · November 8, 2025, 8:49pm

look = display referred

bastibe · November 9, 2025, 5:36am

I find that AgX contrast adds a nice punch of saturation together with the contrast. Which I’m actually somewhat ambivalent about, as one of Darktable’s strengths over other raw developers, was that it had tools that decoupled contrast from saturation. It does look good, though, so I’m not complaining.

But for additional saturation, I found both AgX purity and AgX saturation a bit unnatural, and have gone back to color balance. In particular, the soft roll-off towards the highlights just looks pleasant (although I find the default preset rolloff a bit too strong).

kofa · November 9, 2025, 12:45pm

What I intend to be the final version of the PR is up (fixes, updates based on the reviewers’ feedback may still cause changes, but there’s nothing I want to change/add).

github.com/darktable-org/darktable

Agx UI toe shoulder sliders and primaries rotations

master ← kofa73:agx-ui-toe-shoulder-sliders-and-primaries-rotations

opened 10:35AM - 02 Nov 25 UTC

kofa73

+662 -249

- The primaries hue and purity (incl. reversal) sliders now have visual hints (h…ue, purity); - The hue and purity sliders look like in _color balance rgb_, not like in _sigmoid_ / _rgb primaries_ (there is no white stripe from the neutral position to the selected value, obscuring some of the visual hints) - => it may make sense to alter _sigmoid_ and _rgb primaries_ the same way, but I did not include such a change in this PR - the black/white, toe / shoulder sliders have white/shoulder on the top, black/toe at the bottom - a new 'exposure picker' (_camera_ icon): the module will read the exposure parameters from an early, enabled, preferably unmasked instance of _exposure_, in order to reflect user adjustments + in-camera exposure compensation and 'highlight preservation' modes. This required changes in `_exposure.c_`, `develop.*` and `imageop.*`. Unlike filmic, no estimate (using hard-coded exposure of 0.7 and reading the EXIF) is used, and no adjustment is done in `reload_defaults`, as the exposure instance is not yet available when it is called. The previous method (picking the limits based on image content) also remains. - in the params, the pivot input (x) coordinate is now stored explicitly, instead of storing a shift. The slider has been renamed to _pivot relative exposure_, as it displays the value in EV. The value is maintained if black/white relative exposure is changed via the sliders or exposure pickers. => module version upgrade to v7 + migration The consequences: - the pivot's distance from mid-grey remains constant, therefore if exposure limits are adjusted, the projection of the pivot remains the same (the same input levels are mapped to the same output levels as before). Previously, since the output remained unchanged, but the input changed, the mapping changed, and the region of highest contrast moved. - the shape of the curve will change, reflecting the movement of the pivot _relative to the mid-grey-to-end-of-exposure-range_ distance. - pivot input (x) will still be enforced to stay inside the exposure range; if you shrink that so much that the pivot is forcibly moved (because it would fall outside the exposure limits), then expand the range again, the pivot will **not** be restored. For example: your black and white relative exposures were -8 EV and +5 EV. You set your pivot at mid-grey +2 EV. Then, you set a very low white exposure limit, like 1.5 EV -> the pivot will be forced to 2 EV (minus epsilon). Then, you restore your white relative exposure to 5 EV -> the pivot will remain at 2 EV. - there are now 2 pivot pickers, since some disliked the idea of the pivot input (x) picker also adjusting the output, since that is done by calculating what the brightness would be if we used the default mid-grey -> mid-grey mapping for the pivot. So, the picker next to _pivot relative exposure_ only changes that slider. In order to set both the input and output value (as was done previously), there's a new picker, placed next to _pivot target output_. This may be a bit confusing, at first, but if you think about it, it makes no sense to set the output using a picker without also saying the brightness of what region is to be set to that value. Explanatory tooltips are provided.

Note: the automatic adjustment when opening the module for the first time is not implemented due to technical limitations. You have to click the camera icon to get it.

JasonTheBirder · November 10, 2025, 5:14pm

I still prefer color balance RGB due to the ability to modify purity as well as saturation. The saturation slider IMO is too coarse compared to the fine-grained control of the CBRGB.

JasonTheBirder · November 10, 2025, 5:17pm

Whoa thanks for the info!!! I gotta digest this!

mino · November 10, 2025, 6:49pm

I’ve made a couple of comparisons for adding saturation between AgX and cbrgb. Generally you get more saturation per % in AgX compared to cbrgb and also I feel like AgX adds more “brightness” or “glow”.

hatsnp · November 10, 2025, 7:02pm

That’s interesting, I thought saturation by definition would always equal lower brightness as well, is this not the case for the AgX saturation slider?

priort · November 10, 2025, 7:21pm

Looking at the waveform was more informative to me than trying ot look at the sweep itself…

AGX

RGBCB

kofa · November 10, 2025, 9:04pm

The look|saturation is a simple RGB control: it calculates the Y luminance, and updates the distance of each RGB component from Y according to the saturation setting. It is not guaranteed to preserve Y (in most cases, it will not). Use it for small adjustments only, and use something more sophisticated for precise control. AgX is a tone mapper, not a complete image editor.

github.com/darktable-org/darktable

src/iop/agx.c

08c190245


      
          const float luma = _luminance_from_matrix(pixel_in_out, rendering_to_xyz_transposed);
          
          // saturation
          for_three_channels(k, aligned(pixel_in_out : 16))
          {
            pixel_in_out[k] = luma + sat * (pixel_in_out[k] - luma);
          }

leonidas111 · November 11, 2025, 7:02am

If everything is collapsed, the module takes more place as needed (DT 5.3.0-809, X11):

kofa · November 11, 2025, 7:20am

This has been reported below.

The size of the notebook is determined by the tallest tab, even if not currently visible.

This is a feature of the notebook, and other notebooks behave the same way.