I still struggle with filmic RGB and maybe I did not yet understand.
So far I’ve understood it is to map the range of the sensor from lets say -xEV via 0EV (for middle grey) to + yEV to roughly 0 … 100% display, right?
I hope I’ve understood that correctly otherwise the remaining stuff does not make sense. I know that in theory it is possible to adjust middle grey but lets forget about that detail, since it is a source of further confusion.
Now how to map this? When using a Fuji camera with tone settings 0 0 for highlights and shadows I’d expect the jpg of that Fuji to map -x EV to 0 & and +y EV to 100% if -x … y is the range the camera is capable of. So far ok?
In Darktable using the sliders and observing the histogram I may find out what the range of the sensor approximately may be. One question on this: Can filmic RGB can find out about that out of the picture data? So far I see currently it isn’t done is it? So since it is “physics” of a sensor it would be fine if this needed not being adjusted or being adjusted automatically from looking what is in the sensor respectively the “pixel pipe”, since this EV are actually also affected by darktable itself, aren’t they? Which may also confuse people it this is the case…
Back to the linear conversion -x EV to 0& and +y EV to 100% . What I see in filmic rgb is e.g.:
So a linear curve actually should connect -6 EV , 0% with in this example 2,1EV with 100%, but this cannot be achieved. Why?
I’d expect that I have the creative freedom that I can lay a tone curve on top which allows me to create exactly that curve that I want to have as I can do in all “normal” programs being it gimp or whatever. Of course I still can use a tone curve in dark table but I need to take then into acount that the filmic rgb curve already has a non linear behavior so that whatever curve I lay on top is somehow “bended” with filmic rgb underneath.
Do I understand this correctly or do I overlook something?
I hope it is clear what I mean, otherwise I could provide an example.
Middle grey is the key…so don’t discount it. It is the anchor for the scene-referred workflow and key to the integration of the modules… Your first step after perhaps cropping or whatever your Initial compositions steps might be is exposure… You are going to determine in your scene what value gets mapped to display grey by adding or subtracting exposure. DT by default adds around 0.7 as a reasonable guess but you will often need to change it… I actually use the picker on the whole image set to 50% and its often a good starting point… The image often blows out at this point but if your subject or midtones are nicely exposed that is all you want… Then you apply filmic and map the dynamic range around this anchored point… Filmic actually has both the gamma and a look curve so its not a simple s-curve as you might assume by looking at the display… There is a lot of information about it and i think the manual does a fairly good job of explaining how to use it… THere are also several good videos…
A good balanced coverage from a standpoint of detail and explanation can be found here in a 3 part set
the mapping involved three fixed points: -X EV → 0%, Y EV → 100% and0EV (~0.18) → 50%. As you want continuity at the 0EV/50% point, you cannot use a linear interpolation, but you need at least a 2nd or3rd order polynom.
filmic’s mapping is not directly related to the sensor range, modules before filmic in the pixel pipe can change the covered range. (Extreme case: overuse of color balance RGB highlight brilliance )
you cannot assume that the camera maps the sensor range to 0…100%. I have good reason to believe that my camera (a Sony) sacrifices part of the sensor range to provide more contrasty images with default settings.
In fact, you can’t assume anything about the in-camera processing (well, that’s a bit too strong perhaps…)
Of course you can use an additional tone curve, even when using filmic. If you apply it after you have set filmic, you will see the effect of the curve on the final result, so there’s no need to take filmic into account explicitly… (that would be necessary if you coded a tone curve “by the numbers”, but I think most of us edit by eye).
In this respect, filmic (and sigmoid) behave the same as the “basecurve” module, which also applies a tone mapping (as does any tone curve, including tone equaliser)
the videos, both, the video before and the ne which you mentioned, Martin, I’ve seen of course. Partially even twice not to oversee something which may still be the case.
What I want to do is just to control the gradation.
Here in the example of PS. I still did not find how to see where a certain pint in the picture is in the histogram as gimp does which is sometimes very helpful.
In darktable I understood that the exposure is usually done to lay the intended “middle gray” so that this corresponds to the standard middle gray that the middle gray usually can stay at those standard 18%. But as one see in the darktable picture which I posted, 18% is shown close to 100%. If we’d map e.g. -5EV to 5EV it would be just just in the middle between 0 and a 100%.
Oh by the way I am actually an automatic control engineer knowing about frequency domain and things like that.
It is just my point how to map the way way of thinking from gimp to darktable in an easy way. I found my way of thinking again in PS and I am still looking for it in darktabe…
If you export to gamma 2.2 sRGB JPEG, you’re already using two tone curves: 1) the filmic operator, and 2) the TRC in the sRGB profile.
I sometimes use two tone curves (before the sRGB TRC) for high dynamic range images. I’ll first do a log curve that lifts the low data, then a control point curve to shape the tone to my whim. I’d probably not use filmic for this, as the first curve lifts the shadow data out of the filmic “toe”, which is the distinctive feature of this curve.
The original premise behind a so-called “filmic” curve is one that emulates the response of film, with a non-linear toe on the left, and a gradual rolloff to white at the right. Here’s a link to a presentation put together by the person who coined the term:
I presume by gradation you mean contrast (correct me if I am wrong) Filmic is not about getting the gradation right. There are other tools in the pipeline for that which can included the curve tools. Filmic in my understanding is about recording the wide dynamic range of the film sensor into a compressed range that can be displayed on our computer screens and/or print media. I find that rather than trying to interpret filmics curve display with the bias/expectations that comes from using Photoshop that I find looking at the dynamic range mapping is more informative for my understanding of what filmic is trying to achieve. You can cycle through the graph views.
A couple of issues with this statement. Filmic should not be applied to a JPG, but I don’t believe you were suggesting that it should. Secondly, a jpg is an 8 bit image and would map pure black as 0 and pure white as 255, giving 256 shades of red, green and blue. This can be displayed on a good computer screen. However, not all scenes contain pure black and pure white so the displayed range may be much less. This would be especially true for an image taken on a foggy day when there is only mid-tone shades of grey. So am am not sure how comparing what happens to a JPG image can be much help to understanding filmic.
Good luck with getting the answers you seek about filmic.
Yes, but it’s still mapped correctly, the way the graph looks is somewhat misleading.
Get a grey ramp (e.g. from darktable's filmic FAQ - #29 by kofa), adjust exposure, place a colour picker on mid-grey, adjust filmic: you’ll see that mid-grey is not changed.
You can make it ‘linear’ (actually, logarithmic, if I understand correctly) by tweaking the shoulder/toe on the advanced tab (try all options soft, hard and safe).
Here is an old preset, you may have to tweak if for the current version of filmic:
Wrt. the curve displayed in filmic (“look only”): the horizontal axis is marked in EV. That is an exponential scale, in that “adding 1EV” means “multiply the value by 2”. So a visually big change in negative EV is actually a very small change in absolute value (-5EV relative to 0.18 is 0.18/2^{5} or 0,0056, -6EV would be 0,0028; +2.5 EV would be 1.01)
So a straight line in that graph is not a linear mapping… Which you probably wouldn’t like anyway.
And once again: the filmic mapping curve has to go through three fixed points, and that cannot be done with a straight line in the general case (of course, you can get lucky…)
It sounds like you might have misunderstood to some extent what filmic is for.
Filmic is a “technical” module, existing to address a technical necessity in the image formation process. Specifically the mapping of an HDR signal to the SDR “capabilities” of the output media (screen or print). It does have some creative uses, but it should first and foremost be seen as technical, with the various options existing to help with managing the HDR to SDR mapping in a pleasing manner.
As @paperdigits said, the main tools for what it seems you want to do, are the tone equalizer and the various curve modules.
Currently my feeling is that filmic rgb is to complicated for me right now. Further I have the technical feeling that technical constraints like “avoid negative numbers”, physics and design aspects are mixed up by which it obtains this difficulty also in an objective way.
Thus I am afraid that I am currently to short in time, to come to the point with filmic rgb. Maybe that I will come back in the future.
Maybe in the meantime somebody else comes up with easy recipes how to use it and why to use it with these recipes.
I don’t think its that hard. If you want to alter both sides of your histogram, increase Contrast under the look tab. If you want to alter lights, use White Relative Exposure. If you want to alter darks, use Black relative exposure.
Use filmic only to control your over all tonal range.
You’re missing the forest for the technical trees. Or, ignore the numbers and trust your eyes.
It’s really not complicated. Normally I use auto tune levels and from there adjust white and black exposure to taste, then in the look tab I adjust the contrast if I feel it’s needed and finish by pulling at the highlights saturation mix slider to see what I prefer. Done. Not once do I worry about any of the numbers.
If you were to look back at many videos done by @s7habo and watch he often didn’t even touch filmic… He would add a sweeping 2-3.5 ev to expose the subject and then simply apply filmic with little or no adjustment and then use CC and rgb CB modules to shape the image… I think this is the hardest thing for people coming to DT is to add lots of EV… Doing so often blows out the image and then when adding filmic, rather than just do a basic tweak on the b and w levels and maybe contrast the slider pushing begins to try and get something like the final image. I think it is stated clearly even in the manual and many times elsewhere that filmic is a global tone mapper to manage DNR in the image… Once this gets embraced or maybe people shift to sigmoid which has less options and now the color primaries tweaks it will becomes less of an issue…
Maybe this isn’t helpful and I’m more than happy to be corrected, but I always thought of filmic as mostly a set of guardrails to stop your image breaking. You can move the guardrails somewhat to affect the tonal limits of the image and the compression needed to keep the picture within bounds. But the major style transformations happen elsewhere. There’s other features of the module but that seems, to me, the main one?
I think you kinda got it right. filmic is just another tone curve, designed to reshape the tone distribution of the linear raw data. sigmoid does the same fundamental thing, so does the control-point curve. darktable’s design appears to be “guardrail-enforcing”, discouraging if not keeping one from entering settings that produce an undesirable image (guessing a bit from others’ descriptions, not direct experience).
)
