The exposure range auto pickers (or manual adjustment, does not matter) preserve the contrast around the pivot (not the one you see and adjust, but the one used internally, which makes sure that a given contrast value displayed on the UI remains unchanged in terms of dy you get per 1EV around the pivot). The rest of the parameters don’t change. However, if you adjust the relative exposures, you change the ratio between the number of EVs to the left and right of mid-grey, and therefore move the pivot along the x-axis (you start from 10:6.5, but you may move to 8:3, for example). As long as the change is relatively minor, it won’t change things much.
If you are worried about that, use the toe/shoulder params. I’ll try to improve on those a bit during the weekend, but that will mean current edits will get invalidated, again.
Ok thanks for the explanation. That would definitely mean I’d want to avoid using them if I’ve tweaked the other controls to perfection. I would really want to avoid a dance of the sliders with constant back and forth.
You mentioned making those relative exposure controls “in case of emergency tools”, so do you envisage them moving to the advanced section? I would be in favour of decluttering the main section if the other controls can essentially replicate what those relative exposure controls do.
if needed, lower white relative exposure to fill the histogram/waveform (output dynamic range). I didn’t feel I needed that here.
Here is an alternative, using shoulder start instead. The end result is very similar: we start with the same pivot, with the same contrast, and are very close to white (y = 1) at around 3 EV above mid-grey:
The auto-white delivers a similar result (there is nothing really white in the image, so the auto-picked value has to be increased a bit – see the slightly blown highlights and the overly contrasty overall look).
So, I think the auto relative exposure level picker, and the relative exposure sliders are generally useful tools (not just ‘in case of emergency’ fallbacks).
Today, I ran a few tests with AGX (version 5.1.0+952~g78cc81807f) and, for the first time, I used the “logarithmic tone mapping only” option.
This gave me an idea: combining AGX and Base Curve.
Here are the results below with basic editing.
Using the “smooth | base” preset to start the processing,
We can now use the white relative exposure to fill the histogram upwards. Here the contrasts in highlights are very good, but the white skirt is overexposed:
My conclusion is that the most important thing is how you set the exposure. Pivot x should not be very far off, then you can use it to make fine adjustments.
A new build. Functionality-wise, nothing new (I hope…).
There are some internal changes (simplification of the code). They should not affect functionality, and based on my testing, they do not.
What you will see:
the target black/white sliders are now in percents (hard limits are 0-15% for black, 20-100% for white), just like in filmic rgb and sigmoid. This means, I’ve broken your edits, again: the previous white = 1 is now reported as an invalid parameter, and the new minimum value of 20 will be used instead. However, as that is now interpreted as a percent value, you’ll have to double-click the slider to bring it back to 100%. I did not update the code to use 0.0152% for black (the default black level of filmic rgb and sigmoid). If someone can convince me I should do it, I will.
I’ve fixed a small issue where the black point would not be reached in case of really weird parameters.
The pivot now stays between the set black and white levels, even if you try to push it outside using the pivot y (linear) slider.
I’ve soft limited power values to a minimum of 1.
auto-gamma is now off by default.
The code base has been refreshed with the current master, up to and including g48486b7f.
You also seem to have a few tiny overexposed specks, and loss of contrast – but I think the finest details of the dress are not that important, if the details of the people’s faces are well presented (as is the case in your edit):
Oh, and if you get ‘agx’ has an introspection error, just fix the last history entry, and then compress the history stack (invalid params in previous history items also trigger the warning):
Don’t those ‘Canon-like’ etc. base curve presets assume linear input?
Of course, if they work fine, sure. If I remember correctly, rgb curve shares much of the processing code with base curve. agx itself is based on its own curve and the primaries processing, though, and will remain like that. Without tweaking primaries, you will not get the nice ‘path to white’ that agx provides.
