Defaults in filmic RGB

At every new half-yearly release of darktable, I read the new manual, as a refresher exercise. I noticed a couple of things about the defaults for filmic in 4.2 (I don’t know if this is a recent change or what, as I don’t remember the behaviour in 4.0).

  1. The latitude is set to 0.1% on new edits. Surely this is wrong. Is it read form a config file anywhere (i.e. something wrong with my setup), or is it a bug?
  2. The “contrast in highlights” and “contrast in shadows” are both set to hard for new edits. The manual says the defaults are “safe” (which is what I remember). Is the manual in need of an update, or is this another problem.

The manual is not up to date 100% yet.

The change was intentional – see rework scene-referred defaults · darktable-org/darktable@9a82b2f · GitHub

2 Likes

Thanks Chris.
I knew about the 0.7 exposure default adjustment change.
What is the rational for the near-zero latitude default? It clearly conflicts with the advice in the (presumably yet-to-be-updated) manual of maximising latitude when possible.

Agreed. I hope the author proposes some changes to the manual to cover this, but I’m not holding my breath.

Hi,

With filmic before v6, latitude set two things:

  • the range within which contrast was constant
  • the range where saturation was maintained.

With v6, the point about saturation no longer applies (there is no forced desaturation outside that range).
As for maintaining constant contrast: the wider the constant-contrast range, the more compression you’ll need in the shoulders.

I would think that Aurélien, who added those changes to his fork (Ansel) prefers his filmic to be set that way. For his use cases, those are probably the most convenient settings.

2 Likes

It might make sense in the context of not being needed for the saturation boundary. To me setting it to essentially zero should offer a full/smooth roll-off towards the middle gray from both directions and then people can dial in what they like for latitude?? No idea otherwise if this is not correct… I have pretty much abandoned v6 for v5 anyway so the changes at this point don’t bother me as I use a custom preset… It would likely be good to correct the manual as it would seem that one thing is suggested and another is offered…

Who ever wants to contribute to the manual can, its extremely accessible and is documented. Nobody has enough time to finish all of it this release, so its still in the works.

People ask what they can do to help, well this is it. It requires zero coding skills and helps others a lot. It just requires some time and dilligence.

2 Likes

If I was confident in my understanding of why I would offer to write a short paragraph… would it seem logical to you that since latitude is no longer tied to the saturation curve as in v5 that having latitude at 0 offers a the smoothest possible roll-off towards the middle gray set point and so therefore would accommodate the widest range of images wrt to a default and unbiased starting point??

There was more updated than just filmic…

Hi Todd, I am just wondering what you prefer about V5 over V6. I am just interested in your logic as understanding different viewpoints is very helpful with a complex program like DT.

the answer to this question on french forum by the author himself mid july 2022 :slightly_smiling_face:
"The latitude is the linear zone in the center of the S-curve. The curved parts at the start and end of the latitude have the same slope as the latitude (defined by the contrast), imposed by the algo.

Curved parts can use 3 types of equations. In the interface, they have been popularized by “hard”, “soft”, “safe”. The safe was forced by default in dt because it is the one that never oscillates. But I find that it produces a soft contrast, which people over-compensate with the contrast slider, which only operates on the midtones (with a very marginal effect on the extreme luminances, hence the need to exaggerate contrast – by exploding the midtones).

So the idea was to go back to the hard curve, which keeps midtones soft without sacrificing as much local contrast in the highlights. But the greater the latitude, the more the curve is constrained and the more it increases its probability of oscillating by exceeding the limits of the dynamic range.

So I practically canceled the latitude for safety, which changes the behavior of the midtones very little since the slope of the curved areas is forced to the same value as the slope of the latitude, in order to ensure the smooth connectivity of the portions of curve."

8 Likes

I don’t know the rational behind these changes , so can’t update the manual.

But when i tried r&Darktable for the first time, i noticed i liked the results more out of the box . I started hunting which change was responsible for that, and it was thing one .

I’ve been cherry picking it for my custom builds ever since , since the commit was included into main Darktable (months later)?

So, i like what it does (although i toggle the contrast hard/soft/safe settings often to see if i like one over the other ). It seems to make filmic less super flat out of the box .

So it has been in use since +/- 3.9 Darktable , so i hesitate to call it ‘wrong’. But explain the why and how , i can’t really. Maybe @flannelhead could explain what it does in human talk , although the commit was fine by aurelien on his own in his own fork i think .

Mostly I don’t like the salmon colors and some other results that come from the gamut mapping or hand cuffs as they have been called when using v6. With v5 I feel like I get to decide right or wrong how saturated things will be. I also like the look I can dial in with the saturation curve in conjunction with tweaking the latitude span and location. Just overall I feel like its a more workable version than v6.

2 Likes

Well, these new defaults at least give smoother transitions between shadows → midtones and midtones → highlights, leaving the shadows and highlights more room to spread out the values (i.e. more contrast). With the ”safe” curve and a wide latitude you would often end up with quite sharp transitions at the extremes of the latitude, and highlights and shadows could be crushed without getting much contrast.

I think the mentioned usage pattern (increasing midtone contrast to get more contrast in the highlights and shadows) is a good discovery by Aurélien, and the compensatory change to the defaults is welcome.

1 Like

What about new wording for the docs then?

Feel free to create a PR against the doc repo:

1 Like

I’ve already tried for a much simpler change - it’s refusing to authenticate me when I push (but I can login to github using the web browser). Does my login id (colin-adams) need to be known to dtdocs in some way?

You cannot simply push to the darktable repos. You need to fork the repo, make your change in your own fork, then create the PR.

I agree with the historical, I asked the author after using rdt/ansel for the first time in august.