In my previous post I brought up our major problem,
which is that the foreground and background colors don’t match,
so that even if we can bring their darknesses closer together
the foreground will seem too red and/or the background too
yellow. The question also is how far to go in equalizing the two
halves. Should the foreground seem to be in the same intensity
of lighting as the background, or should we leave it somewhat
darker (although certainly not as dark as in the original).
This image has a lot of challenges but also
opportunities for the use of LAB, particularly in terms of ease
of isolating objects with Blend If. These comments have several
demos that illustrate it.
701 We start off with two versions
faithful to my preferred treatment, that the choir should be
significantly darker than the background. This one, from my 2009
classes, is the better of the two. We know that the gap between
the two halves needs to be reduced. I don’t think that this
person went far enough in lightening the choir, but I don’t see
that the background needs to get any darker. He has managed to
get more separation between the gilded altarpiece and its
surrounds than is even found in the darker par.
702 The choir is excellent. The
common problems with the skin have been avoided. The person
understood that the background needed more attention and added
Vibrance to it, among other things. The real problem, though, is
that it still appears washed out. Compare it to #701 to see how
the background can remain light and still be interesting.
The easy cure here would be a multiplication, as
described in the comments to #706. There is also a mention of
this version at #713.
703 Not the
version most would prefer, but perhaps the most important. This
person did his color-balancing based on the synthesizer, the
steps and the black clothing in the foreground. He then
overwhelmed everything but the choir with a vignette so
heavy-handed that it made this into a night scene.
1) Open your own version.
2) Apply #703 to it at 20% opacity.
I offer no guarantees since I did not test this on
every image, but it dramatically improved every one I tried it
on, including my own—and the par.
The trick works because #703 as it stands portrays a
darkened church with a spotlight on the choir. Spotlights are
very effective in directing our attention to something. We can’t
afford to go as far as this person did, because our audience is
never going to accept it, particularly since they can see
sunlight coming in at upper left. If we blend it into ours at a
low opacity, however, they’ll never detect the spotlight.
Considering that I spent much of Chapter 5 of On the Law of Simultaneous Contrast discussing a
variant of this procedure, I am attributing my failure to use it
to incipient Alzheimer’s.
Also, I should have remembered my own motto: when
everything is colorful then nothing is colorful. That’s one of
the big dangers in this scene. Adding the darkening of #703 at
the edges helps fight that sensation.
704 The first one to present what
I’d call our consensus view. On reviewing this version along
with all the others, I don’t understand why I didn’t make it
part of the par suite. It is among the top in terms of the shape
of the faces and of the dresses. It maintains the choir as
darker than the background. The yellow cast in the background
isn’t eliminated but it’s tolerable. And the person thought of a
useful trick: he used the Channel mixer to spice up the greens.
Apparently I felt that it wasn’t colorful enough to be one of
the best, but I don’t think that now. I wouldn’t mind more color
in the altarpiece and in the faces. The dresses are duller than
in many other versions but the more I look at this exercise the
less I like the idea of brilliant red in the dresses. The idea
should be to highlight the singers, not what they’re wearing.
705 Shadows plugged, highlights
blown, greenness in conductor’s hair.
706 This person paid careful
attention to the foreground, which is good, but left the
background washed out, which is bad. Additionally it is somewhat
colorless. It happens that #707 has a similar treatment of the
background but is generally more colorful. Even though it has a
yellow cast in the background, which #706 does not, I consider
#707 to be preferable. Fortunately there is an easy fix.
1) Add a duplicate layer to #706.
2) Set layer mode to Multiply.
3) Add a layer mask. To it, apply the green channel
from the Merged layer. That is the choice because in the green
channel the dresses are nearly solid and the faces fairly dark,
so in the Merged (multiplied) version the dresses will
definitely be fully black and the faces nearly so. Therefore
when used as a layer mask the dresses will not change and the
faces only a small amount. Meanwhile the background will be
permitted to darken. We could not use the blue channel as a
substitute because although the reds are just as dark there as
in the green channel, the background is also dark.
4) Gaussian blur the layer mask at a Radius of 30
pixels or more. This move is always necessary when blending
layers of drastically different weights.
5) As a trial only, move the file into LAB without
flattening. This sometimes changes appearance enough to prefer
flattening in one space rather than the other. Here, IMHO, the
RGB trial, being less of a bright yellow in the background, is
preferable. Therefore, Command-Z to cancel the move to LAB,
flatten the file and save.
The result is arguably as good as the par. It is
helpful that the dresses are a less obtrusive red. This case
study should remind us (see also #703) that having color
everywhere is often a good thing, but not always. The next
version, #707, is similar to this one but much more colorful. If
we compare the two as posted, I have to prefer #707—this one
looks weak next to it. But if we build up the background as in
the demonstration, then it becomes more austere, more powerful,
and preferable to #707.
707 Another one from my 2009
classes, more in line with our consensus thinking about what is
wanted. It has the yellow cast so many of us do in the
background. The foreground is excellent. See comments on #706
for some thoughts on the importance of strong coloring.
Sutorius says that he believes the photo was taken in Italy, and
therefore he chooses what he considers an Italian red color for
the dresses. This strikes me as a rather deep position to take,
but there you have it. I personally like that the faces are
darker than the par. The dresses are a mess, however, because of
the sudden transitions from red to near-black. Also, there is
too much noise in both the faces and dresses. That would have to
be taken care of somehow but for now I’ll talk about fixing the
dresses with some LAB trickery.
DEMONSTRATION, ignoring the JPEG
artifacts which would have to be removed some other way:
1) Start with the par version. Add a duplicate
2) Apply #708 to the duplicate layer in Darken mode.
Since the choir and foreground in #708 are darker but the
background lighter than the par, we now essentially have the
foreground of #508 married to the par’s church background. This
is a nice improvement in my opinion.
3) Change layer mode to Luminosity. This yields
brighter red at the top of the dresses and warms up the floor.
These are also improvements, so I suppose we could just flatten
and save now. But we still have that ugly posterization where
the dresses jump from a saturated red to almost a black. So,
4) Move the file into LAB, WITHOUT FLATTENING.
5) Flatten the file.
6) Bring it back into RGB.
7) There isn’t any Step 7. We’re done.
How can that possibly work? Why does it make a
difference whether we flatten in RGB or LAB?
Well, normally it doesn’t, but sometimes it does,
often enough that I recommend that every time you have a layered
file (not adjustment layers, that doesn’t work), before
flattening try moving it into LAB to see that whether there’s an
advantage to flattening it there. Most of the time there isn’t
much difference, but here there most certainly is. The
reason is that the LAB file contains an imaginary color.
In the RGB version the Luminosity layer demands
something extremely dark, almost a black, to be coupled with the
bright red color of the par. That can’t be done in RGB, so
things just stay black and closed up. In LAB, however, we can
define such a color, even though it doesn’t exist in real life.
So, the layering structure, when brought into LAB, demands the
imaginary color that is as dark as black but at the same time a
strong red. Once flattened, converting it back to RGB forces a
computation of what that imaginary color should look like. The
difference will get split.
709 Chosen for the par
version. This is what #707 should have been, with slightly
better shaping in the faces and a much better portrayal the
background. David Remington:
I found this to be a difficult
image as well and was not very satisfied with what I came up
with. It looks okay in context, but some parts look pretty
rough. The gold figure in the top left, and the gold in
general, is blocky looking. Several people handled the gold
better. Same for the singers. Not much finesse in mine. I
wanted to add color and contrast so maybe I went a bit too
far. Not sure about my choice of dress color either but it
seems to be within the range of consensus. I could have done a
better job with the window as well. Layered with par in color
mode is mostly an improvement. Better dresses, better gold.
Luminosity mode, par is much smoother. A bigger improvement.
In my opinion, the foreground floor is distracting
and should have been darkened. So an even bigger improvement
would be to blend in #703 at 20% opacity. Also, permit me to
translate “the gold in general is blocky looking”. It means that
there is a whole gang of colorless specular highlights that not
only detract from the golden feel, but also compete with the
silver and with the white candles, which are supposed to be
colorless. With a couple of hours a fine job of eliminating them
is possible. With a couple of minutes, the following
imaginary-color move might be a satisfactory substitute.
1) Add a new blank layer to #707, Normal mode. This
will be the painting layer.
2) Set foreground color to something golden,
agreeable to the background. In the lion exercise I suggested a
ratio of B=4A. This particular background seems rather yellowish
so I’d boost the ratio to 5:1. So, in the Color Picker, specify
something like 75L10a50b.
3) Activate a fairly wide brush tool, Normal mode,
100% opacity. Paint all over the speculars. Don’t paint over the
candles or the silver. It can be done quickly. For example, in
the picture frame, just drag the brush everywhere. At this point
a lot of detail has been wiped out, but we’ll get it back later.
Now, an intermission, for educational purposes only, not part of
3a) Change the painting layer to Color mode. Now,
toggling the layer on and off reveals that things have gotten
worse, not better. The speculars haven’t changed and their
surroundings have lost color variation.
3b) Change mode to LAB, WITHOUT FLATTENING. Now,
suddenly, the speculars fill in. This is another side of the
demonstration at #708, where an imaginary color (a brilliantly
red black) was created in a good cause. Here, while the file
is in RGB, the layering structure is a request to produce in
RGB a color that does not exist in RGB. When it goes into LAB,
it becomes a request for an imaginary color—one that’s as
bright as white, but as golden as the background. On
reconversion to RGB some of the yellowness will hold. End of
preview as to why this method works.
4) Back to our regularly-scheduled program. We are
in RGB, with the painting layer filled with streaks of yellow
and set to Normal mode. Now, convert the file to LAB, WITHOUT
5) Invert the painting layer. Now what has been
painted is a dark blue rather than a gold.
6) Activate Blend If, Lightness, Underlying Layer.
Drag the left slider almost all the way to the right, and the
dark blue painting will start to disappear. Continue adjusting
until the blue barely covers the speculars, then option-click
the slider to split it into two halves. Move them slightly apart
to prevent a rough transition.
7) The purpose of turning the brush strokes dark
blue was obviously to help visualize how to finalize the Blend
If in Step 6. That done, re-invert the painting layer to return
it to gold strokes.
8) Change layer mode to Color.
9) The file must be flattened now while still in
LAB. Returning it unflattened to RGB will cancel these changes.
710 Two separate problems that have
each afflicted several others: 1) faces almost completely
lacking detail; 2) yellow/green cast in the background. The
gilded altarpiece should be quite a warm yellow, maybe something
like B=4A. Instead, it’s almost a pure yellow, the A reading is
711 Chosen for the par
version. Hector Davila posts that he understood that the
faces needed careful retouching. He also did work on the choir
in isolation, as well as on the painting. The background came
out a good golden color, not the yellowish cast that hurt so
many others. The price was that the faces became too pink.
Actually the entire foreground is. As often happens, the cast
can be detected in otherwise irrelevant objects, such as the
Yamaha synthesizer. It measures 66L19a21b, a nice red, which is
712 Bill Theis says he went for
shape in the faces. A couple of these girls look like they have
beards. Originally I had this selected for the par, but even at
only 20% weight, these beards were so offensive that I had to
substitute another version. See further comment on this issue at
713 Similar to #702 in that the
shape in the faces is good and the background is rather light,
but the faces here are grayer. There’s also a contrast issue.
Checking with Auto Tone would reveal that the person should not
have stopped with this version.
714 This person did a lot of right
things on the assumption that this is a single unified picture
with standard lighting throughout. Using the ACR filter, he
neutralized the conductor’s score, and found an appropriate
black point presumably in her dress. After some local moves to
the choir he was left with the typical problem of the faces
being marginally too pink but the background much too yellow.
The altarpiece is in fact slightly green. As we saw in the lion
image, often blending the B into the A can get a better golden
look. This file would also benefit from the multiplication
procedure demonstrated in #706.
1) Add a duplicate layer to #714.
2) Set the layer to Multiply mode.
3) Add a layer mask.
4) Apply to the layer mask the green channel, Merged
layer. This is chosen because in the Merged version, the green
channel is going to be absolutely solid where the dresses are,
so they won’t close up or otherwise change at all.
5) Gaussian Blur the mask, 30 pixel Radius or
thereabouts. This is always necessary when blending to versions
that vary greatly in weight.
6) Convert to LAB, without flattening. As discussed
in #708, we should compare the look of the layered file in RGB
vs. LAB. Usually there’s not much difference, but occasionally
we prefer one to the other and want to flatten in that
colorspace. That’s the case here. When the file goes to LAB the
background gets more colorful, which exaggerates the yellow
7) Command-Z to cancel the transfer to LAB, and
flatten the image while in RGB.
8) Now return to LAB, and proceed to get rid of the
greenness in the altarpiece as follows:
9) Add a duplicate layer.
10) On the duplicate layer, invert the A channel.
This changes the dresses and faces from red to green, among
11) Fortunately it’s easy to exclude the damage with
Blend If sliders. Exclude everything that is not A-negative on
the Underlying layer. Since part of the altarpiece is,
incorrectly, A-negative, it will become A-positive, warmer.
12) Add to the Blend If an exclusion of anything
that is sharply B-negative (either slider will do). This
restores the blue in the top columns. You may also with to
extend the Blend If to prevent certain color changes in the
13) Make a composite layer.
14) Set it to Color mode
15) Apply to it, the B channel, Normal mode. This
adds both yellow and magenta to most of the image, benefitting
both the faces and the background. Reduce opacity to taste; I
thought 20% would be about right.
715 Chosen for the par
version. This person found good facial detail in the blue
channel and blended with it. He didn’t say where, but I think
probably into the green. That got him the good faces he wanted
but had the added benefit of adding shape to the altarpiece and
adding warmth to the background. He might have added to the
effect something like the following:
DEMONSTRATION for advisory purposes
The large painting in the background offers many
opportunities to break up the tediousness of the scene. So, for
the sake of argument:
1) Make a quick rectangular selection of the
interior of the painting (not the frame)
2) Image: Auto Color. This gives the painting more
snap, and makes it much cooler.
Even though attention is not going to be drawn to
the painting, doing something like this further emphasizes the
altarpiece. Of course you would choose a more sophisticated
method, but this one will do in a pinch.
See additional comments about this version at #716
716 This one, done by executing two
versions and blending them 50-50 is perfectly serviceable, quite
comparable to #715. Toggling between the two, however, reveals
that #715 is superior: better shape in the faces, more
definition in the background, less sense of a yellow cast.
Yellow-green everywhere. The steps need to be close to neutral,
but they measure (8)a22b. Windows ditto. Interestingly, the
person noted this and was willing to accept it. Also, even
forgetting the overall cast, the altarpiece is full of specular
highlights which this version makes very dark.
718 Background neutral, steps blue,
faces blue. Apparently the idea is to emphasize the altarpiece.
But this can be done by multiplication or several other
procedures with Blend If, wherefore
Test your LAB/Blend If skills in this way.
1) Duplicate your RGB file and move the duplicate
2) Add a duplicate layer to this file
3) Invert it, so that the background is blue, the
dresses green, etc. The purpose of this is to make it easy to
see what the next step, a Blend If, is excluding.
4) Adjust the Blend If sliders, Underlying Layer, to
try to eliminate almost everything except the altarpiece. For
example, you could start by knocking out anything that is more
than mildly A-positive, which will take care of the faces and
5) Remember that if you can’t get a perfect
selection and there are some purplish pieces elsewhere
afterward, it is easy to add a layer mask, select those
remnants, and fill the layer mask with black.
6) Return to your original. Do Color Boost, Bigger
Hammer, or whatever you like to emphasize the altarpiece,
disregarding what it may do to the rest of the image.
7) When done, convert it to LAB and replace the top
(inverted) layer of the file with it. If successful, that should
limit the move to the altarpiece. And you may find that now you
have differences between the top and bottom layers, you may be
able to refine it by using the This Layer sliders as well.
719 Detail in the faces wiped out.
Too much noise in the dresses. Background seems blurry. The
person says he used a Reduce Noise filter, which comes with a
720 Another one in the category of
“except for one little thing.” Nice color, similar to the par
with the pleasant addition of a better painting and the
debatable addition of more saturated dresses. General contrast
and appearance also excellent. But oh, my, we can’t accept this
much noise in the faces.
721 This is my version from 2009,
which offers a lesson on falling in love with one’s own method.
I was one of those assigned to work on the flat version. I had
recently developed the MMM technique, which the classes weren’t
really familiar with. Once I thought I had the overall color
under control, I wheeled it out, using the red dresses as my
pivot point. I was quite taken with the variation I had put
into the dresses and was thinking that nobody would get as good
a version as this one. When showdown time came I got quite a
rude awakening when it was clear that my brilliant moves had
made the fleshtones and background green.
722 Way too yellow, easily measured
in the skin, the steps, and the synthesizer.
723 Faces are too soft and too pale.
The background walls are darker and less saturated than anyone
else. They do highlight the brilliance of the altarpiece but I
think most would judge it as going too far. The picture has been
transformed from one where the background is much lighter than
the foreground, to the opposite. The impression, I think, is
that the choir has been cut out of a different photo and pasted
into this one.
724 The faces
are pinkish and too pale in context. The background color seems
washed out. A simple darkening of the midtone with curves could
improve this version enough to make it competitive with the par.
See comment at #730
725 Chosen for the par
version. I’ve explained why I think the singers should be
this dark, so it will perhaps not be surprising that this is my
726 For those
who are trying to keep the background lighter than the choir,
this version could be a good choice. The person states that the
correction took him 10 minutes. A believable difference of
darkness between foreground and background has been retained.
The problem, as usual, is that the two don’t match for color.
The background lacks magenta, and the foreground has too much of
727 Gerald Bakker :
Compared to par color-wise, I think my background
is slightly greenish.
Agreed, and the reason is similar to #726 and some
versions of the lion last time. A good B/A ratio for a golden
color is about 4/1. In this version, I have selected and
averaged the lower center of the altarpiece, between the
flowers. The average reading there is 8a59b, or better than 7/1.
Also, while we can’t rely on the colors of the painting to be
predictable, it certainly seems from its fleshtones that there
is an imbalance toward yellow there too. And the long hair of
the woman in the front row measures 50L18a55b, which is
appropriate for a blond, which she is not. The same point in
On the other hand, the gold
ornaments stand out better in my version than in par.
They do, because their color, though too yellow, is
much more saturated than the brownish wall. The par doesn’t have
that, but it does have better detailing.
The par has lighter floor tiles and stairs, but
I'm not sure which to prefer in that respect.
I definitely prefer the darker floor, directing more
attention to the rest of the image.
I found this a hard image to
process. Make the background too light and the golds get
washed out. Make it too dark and it becomes too heavy. Make
the foreground too light and it looks unnatural. Too dark and
the background predominates over the choir.
A very good summary.
Also, there is a lot of
brilliant color, and it's important to emphasize it. But the
pitfall of course is to make everything overly colorful.
All the more reason to incorporate a blend with
But all this discussion hasn’t touched the real
weakness of this version: there is very limited shape in the
face. Some of these girls don’t even have noses. That’s as
unacceptable to me as the noisy faces of #720. It could have
been fixed by better blending into the red, which was done, for
example, by the guy responsible for #715. As that one also has a
darker background, it’s a good choice for blending. But since
we’re coming to the end, we may as well go whole hog to prove
this #727 can be used to produce something better than the par.
So, one final
1) Gather together #703, #715, and the par as well
2) Add a duplicate layer to #727.
3) To it, apply #715 in Lighten mode. This prevents
the background from changing, but replaces the foreground.
4) Add a layer mask.
5) To it, apply the red channel, Merged layer. The
red channel is the lightest of the three as regards the faces,
and the Merged is lighter than the Background layer. That makes
this the best choice to allow the faces of #715 to take over,
with little impact on the rest of #727.
6) Nevertheless, the dresses have gotten somewhat
choppier. To restore them, activate Blend If, and exclude
anything that’s dark in the green channel, Underlying Layer. It
should be easy to find a slider location that excludes the
dresses but not the faces.
7) Change layer mode to Luminosity.
8) Add a new layer. To it, apply the par.
9) Change mode to Hue. So much for the green feeling
in the background.
10) Add another new layer. To it, apply #703.
11) Reduce opacity to 15%.
There are obviously many ways to modify this
procedure, but even as stated above it yields something I’d say
is definitely superior to the par. I’ve posted it as #736.
728 Chosen for the par
version, to which it is somewhat akin and not just because
it avoids the minor mistakes that have plagued others while
keeping a good balance between foreground and background. I’d
like to see more color in the faces and in the altarpiece, but
this is fine just as it stands. Unfortunately, it doesn’t come
from one of us, or even from one of my classes. It’s from
students of Edward Bateman, who explains:
We approached this as an
experiment. They made one version from the flat and one
version from the regular version using Photoshop tools. There
was a bit of back and forth on this as they worked together…
but they did discover that Bigger Hammer (and channel
blending) is their friend. No major conclusions came from the
choices by the class… but they did notice that each attempt
led to their making different choice… in part based on who was
doing the photoshop work under the group advice. The
differences are quite noticeable. In between these two, we
did another version only using the Raw tools. I was pleased
to see that while an improvement, it was the least effective
of their three trials. Finally, to make some comparisons of
their work, all three versions were put on their own layers…
with a blend made of all three at various opacities (with the
Raw tools one adding by far the least.)
IOW, this is sort of a par version of its own. The
accounts for the relative smoothness of the dresses and faces,
without loss of detail. Individual mistakes are swallowed by the
729 This agreeable version combines
some of the best features of #726 and #728. Like #726, I think
it would benefit from a darker midtone.
730 The final representative of the
2009 class, and the only one where the person was required to
work with the open rather than the flat version. It reminds me
of #724, to which I actually prefer it.
731 Admittedly we haven’t been given
the exact color of the dresses but they look rather tomatoey
here, and definitely less saturated than most others. Assuming
for the sake of argument that the dresses in the par are too
assertive, I’d prefer the deeper and less saturated reds of
#706. From the standpoint of contrast, this version is probably
most directly comparable to #712 due to heavy sharpening, but
#712 had better shape in the faces as well as better dress
732 Good color, poor noise
733 Too dark, too yellow.
734 We end with another nice
version, probably most comparable to my #725. Faces too pink,
though, as John Furnes noted in his subsequent posting. He adds:
Having seen the results, I
tried to go all PPW using the recipe for faces 2015 (page 406
second edition LAB), but used Lesser Hammer instead of Velvet
Hammer. The result was much better than my 734, but it would
need some more adjustments. Time spent was less than 5
minutes as opposed to several night hours for the first entry.
735 The par version.