Weird edges after filmic's highlight reconstruct

I’m not an expert using darktable. I only happen to use it because it’s open-source and really feature-rich. Recently, having learned about reconstruction modules such as filmic RGB’s reconstruct and highlight reconstruction, I’ve been able to get some really nice results. However, I also noticed that edges of highlights sometimes have a really harsh red/blue/green border after filmic’s highlight reconstruct is applied. I’ve played around with the settings such as threshold, transition, and even the structure/bloom/grayness sliders, but could never get it to go away. The only thing that somewhat helped sometimes was switching the chrominance preservation mode from “RGB Power Norm” to “Luminance Y”. I’m relatively new to darktable and have never used other editing softwares before this so I would appreciate as simple an explanation as possible.

Another burning question I have would then be, what’s the difference between highlight reconstruction and filmic’s reconstruct tab, and which is better for general purpose reconstruction? Thank you!

Welcome…maybe zoom in and show what you are talking about and if possible provide the raw file in question so as to better allow analysis… I think for the most part the HLR module is set it and forget it unless you are using segmentation and attempting to tweak that… But more commonly you will use filmic and or the tone eq if needed…

The highlight reconstruction module, IMO. I hardly ever use filmic reconstruction… so I can’t help much with that I’m afraid. :slightly_smiling_face:

As far as I understand, the reconstruction options in filmic are not a direct replacement for the highlight reconstruction, but more a additional refinement of it. There is a video by aurelien where he explains the process

1 Like

In particular, I’m talking about the red border on the left edge of the highlight reconstructed image, though if you zoom in to the clouds there seems to be what looks like chromatic aberration there too, which isn’t there without filmic’s reconstruct. I can’t seem to attach the RAW for whatever reason, so I’ve attached more zoomed in screenshots.
With Filmic’s reconstruct:

With Highlight reconstruct:

I have also given Aurelin’s video a watch, though I couldn’t really absorb what he was saying if I’m being honest :frowning:

I’ve never seen such a red border. Maybe you’ve found a bug?

The main difference between highlight reconstruction and filmic’s reconstruct is that filmic is more like blooming the highlights (to hide artefacts and soften transitions, maybe propagate a bit of colour) in demosaicked data, whereas highlight reconstruction, especially with the new modes inpaint opposed and segmentation based, tries to reconstruct clipped raw data.


Thanks for the explanation. I think I kind of get it now. Is it possible that such a red border is an amplification of chromatic aberrations? I did apply the raw chromatic aberration module, but the camera that I shot this with is a Sony A5100 with the kit lens, which I know is pretty old and not great for CA.

I don’t think that could be chromatic aberration from the lens - the way it’s down the hard edge is definitely a digital artifact of some kind. Silly question, but are those images exports or screenshots? Just wondering if it’s definitely a processing issue, or if it could be preview thing (which wouldn’t really matter much). Is the kit lens the 16-50? I have one on my Nex5R and it’s not bad for a kit lens, at least my copy. Not in the same league as a nice prime or something though… :smile:

The images are indeed screenshots, but I think it’s likely a processing issue because JPEG exports of the RAW retain those edges if the quality is high enough. The kit lens is a 16-50mm.

Which HL reconstruction method did you use? Looks like it might have been clip highlights. Make sure to try the others.

Aurélien talks about the purple fringing in this video:

I think this one also has something on it:

The red line down the left edge is either bad data or a bug. Try again to post the RAW here so we can see if it happens for others.

The mentioned video is simply outdated.

The purple in the clouds is definitely not chromatic aberration.

OP, what version of darktable are you using?

I have seen this often. I call it magenta hell.

When working with problems like this in the past I discovered I needed to just lower the clipping threshold in the highlight reconstruction module. But don’t go too far or the saturation in the blue sky would be lost. Easy fix and not an uncommon problem in DT.

I would not use Filmic’s highlight reconstruction on this image. I just noticed opening up an example image of ‘magenta hell’ that the new default inpaint opposed setting in the highlight reconstruction module works better than the previous options and just needs the threshold reduced. So thanks to the developer of inpaint opposed. Great job yet again by the DT team.

1 Like

I believe there are two different issues under discussion here. See screen shot below… I think @kofa (and myself) is referring to (A), thye very fine red line at the edge of the image, while @Terry I think you are referring to (B)?
I understood the OP’s question to be about (A), but there was a reference to (B) as well.

Apologies if I misunderstood.
Edit: for got screenshot:

I don’t know what the red edge is in A…
In B lower the reconstruction threshold so this magenta part disappears.

I used inpaint opposed. Results for HL reconstruction were fine too, the problem only arises with Filmic. I’m finally able to attach the RAW so here you go.

DSC00057.ARW (23.7 MB)

Yes, I was indeed talking about the red border. Solving the magenta problem can easily be done by adjusting thresholds, transitions and grayness sliders in my exp., as mentioned by @Dinobe and @Terry . Though even for this, I’m not so sure if I should be adjusting settings in HL reconstruct or filmic’s reconstruct? Playing with inpaint opposed mode’s threshold doesn’t seem to give much better magenta removal for me.

1 Like

I’m using ver. 4.2.1 on MacOS Ventura 13.2.1. Device uses arm64 architecture if that’s relevant.

My bad. I misunderstood. I will look at the raw file later