3.0 How to get good results automatically?

First of all, congratulations on a very impressive release and thank you for all the work and efforts that has been put into this huge release.

Reading the release notes and watching the video(s), it seems that you are advised to use dt in one of 3 ways: the base curve way, the filmic rgb way or the basic way.

In my opinion it should be easy to get a good result using dt. Just open your photo and you should have good looking image as a starting point. And it should be easy to choose appropriate modules for further editing.

If you choose the base curve way you can select auto-apply base curve and sharpen in the preferences. Now you automatically get a good looking image every time you open a photo.

If you choose the filmic way (which apparently is the way of the future) you are advised to disable the base curve and sharpen in the preferences. When you open your photo you get a dull looking image. You have to manually turn on filmic in auto mode and sharpen the image. How do you do sharpening in the filmic way?

If you choose the basic way i.e. using the new basic module to do simple editing should you then turn off the base curve in the preferences? If yes, once more you get a dull looking image. You have to manually apply the basic module in auto mode to get something good looking.

Maybe you should be able to select to auto apply the base curve, filmic or the basic tool in the preferences depending on which workflow you prefer.

From the release notes I understand that some modules should be used and others not depending on which way you choose. It would be nice to know more about this.

Filmic is designed to work with color balance, the new tone equalizer, rgb curve and the rgb levels modules. But what about the “jungle” of the many, many other dt modules? Which modules should be avoided?

Likewise if you choose to use the base curve should you then avoid the new tone equalizer etc?

2 Likes

I don’t share this opinion. Imo dull image = good image.
It leaves room for futher processing and allows me to give it look i want. If i would want auto processing, i would use camera jpg’s.

1 Like

I use it most of the time as it just works for me. filmicRGB still doesn’t replace basecurve for me on most images. And yes I’m a family without dog shooter who likes to use RAW. Sorry.

I have created a style with basecurce, contrast equalizer, color zones to get a resonable starting point. After import I apply this styles to all my images.
If it doesn’t fit you can always duplicate, discard the history and start from scratch. How nice is this with RAW-developers.

1 Like

darktable is not intended to work automatically, but to provide users a full control over the picture.

“Good looking” is something that needs to be defined clearly. Eastman Kodak have conducted many studies over the past 50 years to define a middle-ground for “good looking” pictures, because film emulsions do at once what is split into steps in darktable, without any control given to the user (except the exposure).

The problem we have in darktable is that any camera manufacturer works under different assumptions, and we don’t know them. For example, if they provided the dynamic range of their picture and their grey point, the input parameters in filmic could be set automatically from the EXIF. But they don’t. So here we are.

By the way, the base curves are reverse-engineered on non-synthetic pictures and contributed by users for what it’s worth. No quality control is performed before including them and some of them are really over-baked.

In my opinion, playing the violin should be easy. Just put your hands on the instrument and you should have a good sounding Paganini as a starting point.

How does that sound ?

I don’t know where the assumption that photography should be easy come from. That’s a craftmanship like any other. Just because computers are involved and technics are hidden under a GUI doesn’t mean you can spare yourself the learning curve. It takes 10-15 years to train a musician, 5-10 years to train a painter, you want to be a photographer in what, 15 s. ?

If this is what you think, what you are looking for is photography for the masses, and that market is covered by Kodak disposable film cameras, iPhones and Fuji OOC JPEGS. darktable is not the right tool for you. It is an instrument to let you express your visual creativity. Not a juke-box to loop-play old tunes.

But that comes at a cost: more advanced editings are pretty much voided by the drawbacks of the base curve (blacks crushing, over-saturation, limited shadows recovery abilities, etc.). I suppose the question you have to ask yourself is : “do you want to go fast or go far ?”.

You don’t get a dull image, you get an image with raised luminance that retains the original colour ratios, assuming you will manage those colour ratios yourself in a way that suits your photographic style, because it’s not the place of the developers to decide for yourself what is a good looking picture.

That’s the next step indeed. But iterative design needs iterations… Stay tuned.

No, you got it wrong.

Tone equalizer, filmic, colour balance, RGB curves and RGB levels are modules using scene-linear RGB encoding to apply transfer functions to your picture, while most of the darktable’s pipe uses Lab.

Scene-linear RGB has 2 wonderful properties :

  • blurs applied in it don’t produce halos, which means the blending masks softening will behave flawlessly,
  • it is possible to reproduce any kind of optical filter effect in a physically-accurate way, with minimum side-effects no matter the intensity of the adjustment.

As a consequence, the linear RGB tools allow you to affect separately the luminance and the chrominance, so editings become more predictable. It is advised to use them together because they form a consistent set of tools intended to work together.

The main drawback of the Lab modules is their usable range of settings is quite small, and you get ugly results real fast when you push the settings a bit far, which means bad model. Also, pushing the luminance in Lab doesn’t hold its promises to keep the chrominance as-is, you will witness weird desaturations with muddy blue colour shifts.

You can still use whatever Lab module you want with those RGB tools. Just be aware that they may mess with the linearity of the signal, so the properties of the scene-linear RGB may be voided after you cross an Lab module. But the pipe has been redesigned so Lab modules come last, so whatever module you use should get its expected input if you don’t reorder the default pipeline.

For those who speak French, I just wrote an article about that the last night : https://darktable.fr/2020/01/darktable-3-rgb-ou-lab-quels-modules-au-secours/ (DeepL does a pretty good job at translating it).

As usual. All the sharpening modules work in Lab anyway (which is not good).

16 Likes

@aurelienpierre Funny is, that Google just fixed the camera app to do first blur and then the tone mapping.


Yes indeed-y! After all my investigation and angst these past few years, I’ve come to this perspective: First, do all the conversions and corrections to get a RGB image that still has the original light measurement relationships (this “linear” thing), and then decide what sort of departures from linear express my desires for the final rendition.

Depending on the scene and how I expose it, I have occasionally ended up with a satisfying rendition that came straight from the linear RGB. Not often, but it happens.

As of today, my default proof processing past this “linear RGB” uses only a filmic curve that does a small midrange lift, but keeps the toe flat. This seems to work for both cameras (D7000 and Z6) for the majority of the images, but there’s always additional work as the situations in which I shoot have a lot of variance in the light, both in single-shot DR and shot-to-shot. I’ve had recent cases where I needed to replace the filmic curve with a control point curve so I could pull the shadows down; filmic (at least my implementation) doesn’t accommodate transfers below the diagonal anywhere on the curve.

If I were to abandon my hack software for one of the “real” applications, right now it’d be a tossup between darktable and PhotoFlow. Thing is, RawTherapee has better demosaic and denoise algorithms, so it still has a space in my consideration. With any of them, it’s the specific control over the process “soup-to-nuts” that would be foremost in my priorities.

3 Likes

So maybe I’m not a complete idiot…

8 Likes

Looking at your contribution I doubt it.
I like the article you wrote. I’m surprised that the vignetting can be fixed with exposure + color balance (I use vignetting often but I don’t like the transition. It’s not natural, hard and too intrusive). I will be back to this article I guess to get most of it. Google translate did a fairly good job.

However I have 2 tricks that I need to rediscover now: channel mixer in soft light - I really like the effect it can produce on skin tones (starting with Kodak Tri-X 400, then blend with softlight and decrease opacity). Second is the tone curve with softlight for portrait. It produces nice warm effect (sometimes and with opacity rather low of course).

1 Like

I just fought my way through the article with the help of deepl. It is to be translated into other languages and nailed to the door of every photographer who uses darktable!
:ok_hand:

4 Likes

I don’t think you need to abandon them, just make sire those modules come at the end of the pipeline, which by default they do.

Aurélien,
I have just finished reading your article (translated) spending more than an hour.
Time unbelievably well spent.
Thank you very, very much for all your efforts, which includes this “illuminating” article.

5 Likes

I too want to thank you for a very informative article - answered a number of questions I had. I used Google translate - did fairly well, and saved it to doc/pdf and included your links to your website. Happy to share a link to it if that’s OK, unless you were planning an english version on the darktable blog.

2 Likes

Thanks ! Sure, you can share, or better copy/paste it on https://hackmd.io/ so people can improve the translation and I can proof-read.

2 Likes

Yes, hackmd was what I was going to do… I’d be super happy to copy edit the English version.

1 Like

Thanks Mica - I’m at work and just heading into meetings. Here are the links - would you mind grabbing and posting to hackmd? I haven’t used that site before. And apologies my version is in .doc format and didn’t use opensource!

Fully translated version: https://hackmd.io/Wd0kNkg9Snq1lfC4KnUSlg?view

2 Likes

Here is the article translated to English: https://hackmd.io/Wd0kNkg9Snq1lfC4KnUSlg?both

1 Like

I did not yet read the text very carefully but what do you suggest using instead of contrast brightness saturation in order to increase saturation?

Colorbalance, 2nd slider :wink:

2 Likes

I could help proofread if there is any need. Not sure how much time I have though.

ew the Google translation is wonky