I’m trying to create my Filmic RGB workflow and I defined a my preset for Filmic RGB. My camera has roughly 13EV dynamic range, so I have defined that my white and black relative exposure is ±6.50. As a result I have to increase my exposure in exposure module 1EV – 2EV, even more to get the histogram pushed to the right. I understand that the middle gray is not in the middle and at the end the relative exposures are not symmetrical.
Have I understood something wrong and should I use different default white and black values to get the Filmic RGB roughly right from the beginning.
I think you can’t rely on a generic filmic preset because it is a scene related tool. That is, each scene depiction will have its own specificities regarding those parameters.
However, there are a couple of generic presets that come bundled with filmic.
I prefer to start from neutral (18,45), any white, any black, and contrast 1).
TBH - just go and watch @anon41087856’s videos on it + @s7habo’s recent videos with linear workflow.
Generally - as @gadolf said - it’s not “camera related” but “scene related”.
If I remember correctly - yup: gray to 18.45 % black & white to appropiate spacing, contrast to 1 and have fun.
you can check relevant section from here:
and this is interesting bit about setting for most “range” of adjustments:
I have minimal filmic style that I use to replace not applying a basecurve by default. It just enables exposure in auto mode (this shifts the average exposure to middle gray) and filmic with the 7EV preset. Base_Filmic min.dtstyle (975 Bytes)
Of course this is just a posible starting point, in this case to give you a rough idea of what the picture can look like including the non-linear conversion. Some people prefer not to apply filmic and evaluate the image directly on linear space. In any case, I think the important point is to be consistent.
In rawproc, my default processing includes a filmic tone operator, but I’m finding it’s really only to have the tool in the chain for subsequent tweaking. The needs of the scene are really specific to the scene, it would seem…
After yesterday’s fun trying to accommodate subjects in the shadow of a barn backlit by the pre-twilight sky, I’m thinking of a U3 setting that enables the two-image HDR mode of the camera.
What’s the difference between middle grey at 18 and at 18.45 ? It seems most people reference it to 18 %, but when I look up middle grey I stumble across several tones based on which colour space they’re defined.
They do shit almost 50% of the time. The biggest problem of filmic is it uses a logarithm, which goes to - infinity when your RGB values are close to 0. But 0 is a valid RGB value, and Rawspeed is scaling RGB values between 0 and 1, by offsetting the RGB values with a detection of the dark current on the non-exposed pixels, at the periphery of the sensor.
The logarithm is very desirable for what we do, since it aggressively pushes low lights. The problem is you have to clip the dynamic range somewhere, because an infinite DR is not an option. The question is “where to clip ?”, and “based on what ?”.
So you could naively think that clipping to the camera sensor DR is a good start. But the RGB code values have been massaged already, scaled and offset, possibly more than once, when they enter filmic. Because we are near the end of the pixelpipe.
Filmic OCIO does it the other way around. It’s a collection of pre-computed LUTs, so the filmic curve is hard-coded in different variants (more or less dynamic range × more or less contrasted intent), then your job is to adjust the exposure and black level before filmic because it wont change.
Anyway, there is a style with most of the useful sweeteners tuned by me, apply it, adjust exposure centile and black so everything fits in inside filmic’s DR, and that’s the most automated image processing I can offer you so far : Auto-filmic - Aurélien Pierre.dtstyle (2,0 Ko) . The next step is simply to hire me to shoot your pictures…
Yes, the autos can fail hard. Every time I follow a filmic workflow it is fast and easy to do per image. But maybe I am in the minority considering the hullabaloo.
Thank you all for your comments and styles. The current COVID has reshaped my schedule and I have been only sporadically following the discussions here.
I noticed an interesting thing with @anon41087856’s style I uploaded and have used. When I deploy the style Filmic RGB looks normal. If I change white or black relative exposure, the curve clips even if I revert the parameters back. If I change the values in Look tab, this abrupt clipping doesn’t happen and I can also revert the clipping by readjusting the parameters back.
The same behavior is on Windows 10 with the latest official 3.0.1 and in Linux with the latest 3.1.0 version from git.
I’ve taken to leaving all tonal curves (filmic rgb, tone curve, base curve) off to begin with, and only applying it once the other adjustments are complete (white balance, color balance, tone equaliser, local contrast, contrast equaliser). Then all I have to do is adjust the ‘white relative exposure’ slider in such a way to prevent clipping in the histogram. ‘black relative exposure’ and ‘dynamic range scaling’ can also be adjusted to taste, but are of less importance. Even then, filmic rgb is not best every time. I snapshot my filmic rgb, and compare it to a base curve and tone curve, and keep whichever of the three looks best. It’s the best approach I’ve found so far.
I do much the same. I apply auto set standards for all of those same modules and then use use the middle-grey eye-dropper on filmic-RGB. With most of my images this puts me in a close position for the final tweaking.
My comment here was not how to get the best image, but the behaviour of the filmic module. If I change the white/black distance in @anon41087856’s preset and reset it back the filmic curve is not as it was in the beginning.
My preset changes the display power function (in display tab) to lower the display middle grey point (represented as the big orange dot on the graph — see on the left screenshot it’s lower than the diagonal line, but straight on it on the right).
The reason is the display power in the preset is computed so the contrast setting (in look tab) doesn’t add or remove any contrast in the latitude part of the image (contrast = 1.62 <=> power function = 2.04 → that’s the Dmax value of modern color film). In this setting, filmic barely touches lowlights and only act as an highlights roll-off, which makes it easier to use it hard-set.
If you reset the module, the display power is adjusted automatically whenever you change the scene parameters such that the middle grey stays on the diagonal line. That behaviour will be optional in filmic v4. It is only done on the GUI side anyway, so if you are done tweaking the scene params, you can then adjust the display power later without auto-correction.
Notice that the display power function is equivalent to the paper grade, if you used to print in analog darkroom. I figured that while coding the film inversion in negadoctor.
Ok, if you understand nothing to the numbery part, just know the values come from the light sensitivity of modern film emulsions, through some conversions.
I always throw the maths because I know some people here enjoy them, I should remember to provide the simplified explanation too.
