Darktable/Rawspeed clipping more highlights than other Software


I am currently in the process of making Darktable my main image processing program. Now I came across an old (and quite bad) image with clipped highlights in the clouds in the sky. I didn’t think much of it, but then out of curiosity I opened the same file in On1 (evaluating trial) and was very surprised that many more highlights were present there. Then I tried RawTherapee and the result was the same as with Darktable.

I suppose that some data about my camera could be improved in Darktable/Rawspeed. Is there anything I can do about it or to provide assistance in any way?

The camera is a Pentax K-5. Darktable is version 3.8.1, Windows build downloaded from the Website.

Here are some screenshots. File opened in Darktable with exposure turned way down:

With raw clipping indicator:



I can also attach the Raw file of course, but since i don’t know the forum rules I don’t wan’t to spam with big files.

Go ahead and attach the raws. Add a CC or other permissive licence if you want feedback, though. You could also turn this into a PlayRaw, where people post their processed versions along with the processing parameters, so you can learn from their work.

If I remember correctly, darktable’s raw clipping indicator is not really raw and takes white balance multipliers into account. Might be outdated info, though.
If the image has no or little talk raw clipping, you can simply disable the highlight reconstruction module and recover highlights in filmic, or in some other module.

Ok thanks, here is the raw file. Feel free to use it in any way you like.

I tried all the settings in Filmic, also alternatively in the base curve, but nothing makes a difference.

PXK55990.DNG (21.5 MB)

Please add a licence that allows others to process your image and share the result (‘derivative work’).

Nope, that’s still true, and easy to test: take an image where just the red channel shows as clipped and disable the white balance module… One of the cases where you’re better off not using the highlight clipping module (as it uses the raw white and black points after the white balance module is applied).

Same could be true if blue and red are clipped, but that’s less common I think (red shows easily as clipped, as it has the highest multiplier in the white balance module, and it’s is a common colour component in flowers)

You can get a better result than you showed with darktable , but you have a problem in that image, in that part of the sky is overexposed in all three channels. So there is no detail left in that (smallish) zone top center…
(why are your edits so dark, btw? It makes it harder to judge the sky).

To get some detail back in the sky, I used the highlight recovery module in “reconstruct in Lch” and the reconstruction in filmic. Carefully adjust the relevant sliders to get the best result. (I can’t show my edit here due to licensing restrictions… It’s why you were asked to add a license)

How do I add a license? You can consider the image to be in the public domain.

(why are your edits so dark, btw? It makes it harder to judge the sky).

I just reduced the exposure drastically to make it clear that the clipping wasn’t due to overexposed settings in Darktable…

You can get a better result than you showed with darktable , but you have a problem in that image, in that part of the sky is overexposed in all three channels. So there is no detail left in that (smallish) zone top center…

Yes, but there seems to be much more detail in most of the sky which Darktable seems unable to recover.

Darktable and Rawtherapee don’t clip more highlights , not at all.

If anything they allow you to extract more .

But sometimes it makes you work harder for it , or it requires some tricks and knowhow.

If you open the raw , disable filmic and highlights recovery and just move the exposure slider down, you must get at least the same kind of detail as others .

The highlight reconstruction module is meant to fix parts where one or two channels are clipped , but not all three . Rawtherapee has a similar thing to enable in the exposure tab.
Other - commercial - programs do this sort of automatically, but don’t give you any control over it .

If all 3 channels are clipped (or the reconstruction gives unwanted results ) there is NO program they can magically recover those highlights.
What you can do is try to fill them in with clever algorithms . Instead of 'revealing detail’s you are then 'making up new detail '. Filmics highlight recovery is meant dor that, and for smaller patches .

Of there are large parts clipped with 3 channels , you are often left with the only choices being to ‘blow it out and accept it’, or to replace that part of the image all together.


But I did that. Disabling filmic, using base curve, not reconstructing highlights etc. I never get any more details in the sky than on the screenshot on top. And look at the results from the other two programs. That’s not looking like any clever trickery reconstruction, but they seem to be actually “seeing” more data than Darktable.

Also, it’s only Darktable in this case. Rawtherapee gets the same results as On1.

What I suspect is that the white point for the Pentax K-5 raw (in-camera DNG, btw) is wrong.

Good morning, @firefrorefiddle, and welcome!

a) Here you will find more info on how to “add a license”: [PlayRaw] Waiting for the Big Boy - #4 by pphoto

b) @rvietor’s advice (above) is a very good start.

Of course dt as well as RT can squeeze much more info from the sky, here are three very rough, swift examples:


dt 2 with more oompf (and by using clever masking, you can restrict the oompf to the sky):

Have fun!
Claes in Lund, Sweden

1 Like

Thanks, Claes!

Hm… I stand corrected. The details are actually there. But how did you do it?

I assume, that you have Scene referred workflow activated. Then it’s most likely filmic’s behaviour, that’s causing the clipping. That’s not a problem through, because you can always change that by adjusting white relative exposure. If uncertain, then you can test for clipping in RAW data in darktable.

Hoping, this is of any help.


Start with rvietor’s advice.
To achieve the oompf, use
the local contrast module.
Play around with the controls.
In my case, I used

Note: This will make all flowers and trees go crazy,
of course, so to get a more “natural” look, you have
to mask for the sky.

Play around & have fun!

Claes in Lund, Sweden

PXK55990.DNG.xmp (16.2 KB)

I tried to extract texture, then used graduated density to copy colours into the blown region.
darktable 3.9 from darktable windows insider program 4/3


It looks reasonable, at least for the RPU K-5 DNG sample:

Exif.SubImage1.WhiteLevel                     16316

Could increase it all the way to 16383 (14-bit max) in the rawprepare module, but doubt it’ll make a significant difference…

Thanks Istvan,

you did great! I am happy to look at your XMP file, it’s very instructive. I see now that my original suspicion was totally wrong and the issue was actually caused by the highlight reconstruction module (which I wasn’t aware that it was on by default). Also, I learned about the grey vs colored highlight reconstruction in filmic which is very useful.

Overall I have to say that I am really happy with darktable, the logic it uses and the results it can produce.

And thanks kmilos for looking that one up!


I said to disable filmic to check what’s in the file. Not to enable base curve.

Having nothing activated but white balance and exposure , you should be able to move the exposure slider freely to check where your data stops and starts.

A white point misread from the raw file might actually be a possibility, but I doubt it.

The problem was not caused filmic, but by the highlights reconstruction module. I didn’t know that it was on, and turning it off or changing the method instantly showed that the details are there.

Note that in many cases you can use the highlight reconstruction module in reconstruct in LCh mode; it gives much better results than the default clip highlights mode; on the other hand, reconstruct color is prone to artefacts.

The way I shoot now is to underexpose in order to protect the highlights, then let the exposure module undo my in-camera exposure correction. Of course, I’ll let the Sun, lamps and specular highlights be overexposed. :slight_smile:

Maybe check out this thread for many approaches to fixing highlights:


You will also notice if you use filmic that the chosen norm can really impact the highlights…so you have to take that into consideration.