highlight reconstruction issue

Can you describe what you did? Not everybody has dt git master. I think you used the color calibration module to make the sky look more blue. Because if you only use highlights reconstruction lch you just make the sky grey, but here it is blue. I tried to open the xmp but dt 3.4 does not do this properly.

The shoulder highlight is unnaturally blue/gray and more noticeable than the knee IMHO.

2 Likes

I don’t know about the current iteration of filmic but I recall lights and darks being desaturated because they “aren’t supposed to be” saturated.

3 Likes

_5290185.orf.xmp (8.5 KB)

darktable 3.4.0

4 Likes

i tried it. and yes the color calibration module made it more blue but its a monochromatic blue instead of blue and white.

yes, those clip areas do not have nice transition. i have tried both reconstruct LCH and reconstruct color sometimes LCH is better and other Reconstruct color is better. but reconstruct colors will always have a mosaic pattern what ever you do.

yes, that’s how filmicrgb works, I think. but there are times when you don’t want that

@wpferguson most of the images that i pass through filmic seem to lose contrast. It would take another module such as the color balance module to bring back the contrast. and the recovered sky seem to be much grayer. I feel the best showed here is the one by photoflow. It looks the most natural and the contrast is abotu right. I understand everyone will have a different taste here but the photoflow version seem to be the best to me. even better than capture one and its clean

Had a go at the sky but limited success. Instead concentrated on the subject! dt 3.5, filmic, tweaks. @Nilvus, how do you use the history from your jpeg pls? - Load Sidecar only shows XMP files and the manual isn’t explicit.

2 Likes

You’re right, I don’t check all part in details when it’s not my images.

I’m quite sure XMP inside JPG could be used with darktable 3.4.1 as not so much things related to what I use have been updated since. Yes, I use color calibration. reconstruct highlight with Lch mode as I said and default setting (1.000). Main other things are tweaked with filmic rgb.

To load an XMP from JPG, just use same process like to load an XMP. In the open dialog window, just change displayed files from XMP to all files (bottom right) and then select and load JPG files.

1 Like


looks like this if I try to load your settings in dt 3.4. Color calibration is disabled after an error.

Thanks (doh!)

This is my take with DT 3.5 git master


_5290185.orf.xmp (17.6 KB)

I think we can all agree there is an issue with the highlight reconstruction module reconstruct color. Lightroom implementation is just keep it white if it’s not able to reconstruct would be better than a magenta cast. If there is a way to implement photoflow highlight reconstruction that would be awesome. Need to do more research

There’s an old saying “you can’t make a silk purse out of a sow’s ear” and I think if the sky is important in a picture, then it should be exposed for properly.

7 Likes

Hmm, quite difficult to answer as you don’t show (nor describe it) the error in color calibration, in your screenshot.

If 2 channels out of 3 are clipped there is no color information whatsoever. Color can be propagated from adjacent areas, but there is a limit and it can generate artifacts.
Lightroom is right to reconstruct grey in that case.
In this specific picture I masked the sky area to make it kind of blue, and if it is too magenta it can be changed. But it’s pure invention of the user anyway.
Unless the SW applies some AI black magic.

1 Like

On dt 3.4.1, I didn’t experience any problems with mosaic patterns (not that I noticed), but I just couldn’t find a good solution for the highlight on her shoulder.


_5290185.orf.xmp (142.6 KB)

the error is “channelmixer version mismatch”

At some point, you might want to consider inpainting the blown area.

I drew a mask and used the color reconstruction module