Filmic v6 and white balance interactions

[@Giles29]

So many things in Darktable I don't really know what they do. Always great to learn.

[@emrg777]

Thanks as always for your video...provides better context for the maxrgb norm and how to handle it....

[@luisarevalo6112]

Thank you and team for the great work you all do. I've been using Darktable for almost 3 years and have learned a lot about digital processing and how my cameras affects what I do. This video has a lot for me to unpackage, all very usefull. I use ExpoDisk for WB but still rely on postprocessing for the work I do, my daughter's jewelry designs/jewelry models, and having options for highlight reconstructing and other features are a big plus! Again Thank You and Team!

[@ViktorPankraz]

Thanks Aurélien, for this explaination. I'm struggling with highlight reconstruction in darktable and until now I was not able to get the same pleasant result as in RawTherapee (ART). With all these information and guidance I will give it another try because I like the workflow of darktable so much.

[@philsmith774]

This is an excellent video and I've learnt loads from it. Thank you.

[@Slave-Of-Christ]

Excellent!

[@sherab2078]

I've just started getting to know DT (v. 4.8.1) and I love some tools and features. Especially masking and multiple instances of modules are great things. But I must admit that after years of using RT, I find the colour and highlights management in DT complicated and somewhat unintuitive. I know this is probably on me, but I really struggle to get the effect I would like when in RT this comes somewhat easier to me.

[@pedrorrodriguez1]

Fantastic work! Thank you! Could you provide a bit more insight into the other norms (luminance Y, RGB power, and RGB euclidean) and when to favor one over the other?

There is no norm that is absolutely better than the others. It's mostly a matter of look and preference. The luminance Y, then power norm, are the 2 most contrasty, which is not something you always want. The euclidean is often the one that best preserves local contrast in highlights, but again, smoothing them is more consistent with human vision.

[@pedrorrodriguez1]

@ Thank you! Lately I've been using the euclidean one, so I was wondering if there was any reason besides personal preference to choose one over the other. The manual also says that it is the closest one to color film, which is not a bad thing to me

[@ColinPittendrigh]

Does Ansel have a cli? Command Line Interface..........for converting batches of RAWs according to one xmp file? I have not found one.

I wish there was a setting that as a first approach just clips and desaturates the highlights with a smooth (maybe adjustable) roll-off, similar to what analog film does. It's nice to have all the options to tinker with and be quite forensic about colors, but in most cases blown white highlights are OK. As you say: Garbage in, garbage out. If the highlight is blown, why “invent” colors with time-consuming machine-learning? I see that I spend waaaay to much time tinkering with small overexposed parts of the image just to end up clipping them as elegantly as possible. Sure, there are cases where every bit of information counts, but in 99% of all cases a nice rolloff without any pink or cyan casts are the best and quickest solution for me. In travel, documentary and nature photography there is often not enough time to adjust everything perfectly, skies are white sometimes. So I ended up setting preserve chrominance per default to no and I worry a lot less about the highlights. The other parts of the image are much more important. (Not meant as a rant, I appreciate all the hard work and effort and I am happy to have all the knobs to tune when it's necessary. But most times my feeling is that highlight recovery is mostly an academic adventure for edge cases and do not help me that much as a user.)

[@7xppgngrmj22jhx7enubvnw5grndiy]

I saw that the old "reconstruct color" mode in the same module (highlight reconstruction) often works just fine, without leaving as much magenta artifacts as the new algorithm does (guided laplacians). I bet it could work on this image too.

That's actually not true. On event/shows photography, when artists have a glowing skin lit by spotlights and the clipped areas are on skin, the guided laplacian is the only one that recovers the proper skin color and the smoothness of the face. It really depends on what is clipped, but in general (non-white clipped regions), the older methods are really unreliable. In any case, even with remaining magenta, the guided laplacian is the only one that reconstructs the texture. So if, after this first step, the picture has still a wrong color but consistent gradients (so the clipped region is smoothly connected to the valid region), that's a sane base for further color reconstruction (in filmic or simply by desaturation). The other methods will yield a flat color or a flat "white" (that may not even by white since white balance is not done).

[@ilmangus]

Very nice video, I was just working on images with this kind of situation in the clipped highlight. A couple of questions: 1- I use an x-trans camera. What method in highlight reconstruction module do you suggest to use in a case like this? 2- Highlight recontruction module has to be used only when we have clipped highlights from the raw, otherwise is better to disable it and use the reconstruction section within filmic, am I correct? 3- What about the "luminance Y" chrominance preservation? I noticed that that it works well in some difficult highlights situation (even the non-clipped ones). Does it have some specific caveat that I'm missing? Usually, I use the "cheap trick" of white relative exposure to clip the actually raw-clipped areas and most of the times it's quick and provides good results overall. PS: I must say, your tutorials are getting really better over time. Kudos to you and thank for all your work!

Unfortunately, X-Trans sensors are challenging and there is no method that works really well. The less bad would be the Lch reconstruction. If you have nothing clipped in the picture, anyway the highlights reconstruction is disabled by its the threshold. Same in filmic reconstruction. The luminance Y is slightly more perceptual than the others, there is no caveat. Use what works best. Thanks !

[@StudioPetrikas]

Great video, thanks for clearing a few things up. I had a lot of people ask, whether they should use "Highlight reconstruction" together with "Filmic rgb". The documentation said that it's best so separate them (use filmic reconstruction with filmic rgb; highlight reconstruction otherwise). Another point, that I'm sure people will cling to is that the saturated highlights are not always desirable. I.e. when going for a film-like aesthetic, which sports a very gradual, smooth ""desaturation"". "maxRGB" does a great, great job of preserving the RGB ratios at all costs, but then it becomes difficult to get the smooth desaturation as we reach the bright areas of the image. A good example would be the red flourish on the second rider, next to the white shirt. I think the red becomes 'uncanny', as in, too chroma-laden compared to its surroundings and it feels like it's breaking the grayness boundary as well? Interested in hearing your thoughts. Thanks again for the video, I will make sure to link it every time this question arises.

I think people overestimate the amount of desaturation film applies. I have Portra 400 negatives with blue, blue sky in background that the typical RGB curve would desaturate. The desaturation that film does is contained mostly in specular highlights or emissive regions, it's a bleaching that kicks in at much higher luminance values. RGB curves start desaturating at too low emissions. Regarding the red, I wasn't there, so I can't comment. It seems that the guy in the back has newer clothes than the one in the front. Given that the greens, the other reds and especially the skin tones hold, that might just be what it is.

[@StudioPetrikas]

@ depends on the film perhaps? I've seen some pretty extreme examples (I'd show, but, you know, it's youtube comments). Especially in some of the older (maybe cheaper?) stock. There are often gradients in the sky that fade to white completely, sometimes induced by vignetting.

@@StudioPetrikas It also depends how you expose it and develop it. In any case, the highlights saturation is better managed through color balance than through any tone curve that changes saturation as a side effect of contrast.

[@StudioPetrikas]

@ good point!

[@StudioPetrikas]

@ However, that might lead to incredible fuckups while "manually desaturating", like gray gradients where there still should be colour.

[@cedricbarret4245]

Salut Aurelien. Je viens de télécharger la version 4.0 et j'ai du mal à comprendre filmic V6 + reconstruction des hautes lumières en guide Laplacien. Pour l'instant, je serai en Filmic 6 et préservation de la chrominance sur non. Hate de tester ton fork de Darktable.

Je donne des cours particuliers du mercredi au vendredi, de 15h à 20h, sinon.

[@cedricbarret4245]

@ cours particuliers difficile pour moi. Je serai plus Interressant si tu as une formation payante du logiciel avec des vidéos que l'on peut consulter quand on veut.

[@nikolas01123]

I'd just get it on gimp and fix it there with a color layer.

Yeah no. That's not only inefficient, but Gimp doesn't do scene-referred anyway. Or you will need to export a 32 bits float TIFF of 200 MB to do it. It's just wasteful all around.

[@nikolas01123]

@ Thanks for the reply. It's a good chance for me to thank you for Darktable. I did not mean instead of Darktable of course. Since you mentioned it, is 32bits float a waste in your opinion? I am asking cause I export that way trying to get as much as possible out of these exports.

@@nikolas01123 32 bits float is what you need to exchange data between software if you want to process the image further. The problem is the files are super heavy and, for a finished image, 16 bits are already more than enough (remember the raw file is 12-14 bits). So if you can avoid to save an intermediate 32 bits file, just do it.

[@nikolas01123]

@ Thank you. I'd also like to state that I am disappointed for the route darktable decided to take. With you 100% on this.