What do you mean 'settings are established using one image'?
You use the image to determine where to set the point. For example, by
sampling a gray point or averaging out an area of interest, etc.
Typically in a curves window, you can just select a point on the
curve, and move it up or down. The uncorrected image does not appear as
input.
Yes, you can set points directly (i.e. blind) in the curve but that's
a very blunt instrument. Doing that means making assumptions about the
image. For example, using the "standard" S curve (64,128,192) assumes
where the shadows, midtones and highlights are.
However, even in that case, basing this "blind" setting on a gamma 2.2
image while the curve is applied to a gamma 1.0 image doesn't get
around that root problem.
There is a second method, where you select a point on the image, and the
value of that point gets transferred to the curves window.
Yes, that's what I'm talking about.
It is possible that
Vuescan fails to apply gamma in this case, but that is just a one line fix.
It wouldn't be the first time that Vuescan fails to do the simplest of
things which is precisely why it can be justifiably described as buggy
and unreliable.
If you select bin 128 in the window, then yes, you may end up changing bin
56 in the 2.2 gamma corrected image. But the 2.2 gamma correct image is only
there to display what you are doing. The real image is in gamma 1.0.
Which is precisely the root of the problem, apparently.
You 'cannot' display a gamma 1.0 image directly.
I notice quotes around 'cannot' which means you know what I'm going to
say (so just ignore the next bit) but for the benefit of kibitzers:
Actually, one can. That's what linear editing is all about. One simply
recalibrates the monitor to gamma 1.0 as well and then the image does
not appear "dark and contrasty".
So, in all likelihood,
when you select 128 the right thing is going happen: the curve (with 128 moved)
gets applied to the linear gamma image. Then a gamma of 2.2 is applied and the
results is shown on the screen.
Of course, if that were the case, then yes. But that's not what was
reported.
But even in that case looking at one image while editing another (in
the current context) is just asking for trouble. But such convoluted
contraptions are one of the major trademarks of Vuescan and why it is
so buggy and unreliable.
Don.