Topics

Mantillas: Dan's comments


Dan Margulis
 

After dipping out toes into the Hotel Lobby image, we are thrown into the deep water of this one. It’s no accident that the two follow each other: neither depends heavily on PPW, and in both the best versions rely on channel blending.

This one is, as we all now know, more difficult, because it has more deficiencies and more ways to foul up. There was no real color issue in Hotel Lobby but there surely is here, and the luminosity problems are much worse, too.

Because my comments on the individual versions are lengthy, I’m putting them in a separate post. This one is the introduction only.

Two objects are of great importance here: the women themselves, and what they are wearing. The original gives us plenty of detail in the clothing, so getting it to our taste should not be difficult, unless we make it the wrong color.

The women are another story, and it’s a sad one. Their faces start out very flat, too light, and with little variation. In the event that we solve these problems, as a special bonus we will be confronted with noise and other artifacts.

We must also recognize that these women are from Spain, not Norway. This point escaped a regrettably high percentage of us, who submitted the kind of light, pink skin often found in arctic latitudes.

Complicating things, no bright colors are to be found anywhere in the composition. And all indications are that this is an overcast day.

There is also the artistic question of how dark to make the clothing and how it relates to the darkness of the skin.

How then, should the exercise be approached?

With the clothing, you merely have to avoid doing anything bad. With the faces, you have to try to do something good, because they’re quite inadequate in the original both for color and contrast. Unfortunately, most attempts to improve the faces bring out a lot of noise that needs to be addressed with painting tools. Probably, it’s best to do some shaping also, perhaps using the dodge and burn tools. Special attention is going to have to be paid to the eyes, eyebrows and teeth. It’s easy to lose their definition.

We should know in advance that manual retouching will be needed, but still the best way to add shape initially is with channel blending. In the Hotel Lobby study, we liked the structure of the red channel because it separated the plant from the lighter wall. So many of us blended it into the green channel in Luminosity mode, with some variations on that theme.

But there wasn’t anything wrong with the green channel as such, we just wanted to change the distribution of its objects. In this Mantillas exercise, OTOH, the original red channel is, (to use the technical term preferred by professional retouchers) a POS. Somehow, when the fun is over, it’s going to have to be much, much better than it starts out, or the faces will lack depth. If you don’t believe it, you might want to compare the red channels of the par and the original versions.

How to accomplish that is unclear, although the most common approach was to start by replacing the red with the green on a Luminosity layer. Some also tried the blue channel, which has excellent shape in the faces, as a blend source. But it is rather noisy, and lacks detail in the eyes and eyebrows, so a certain amount of juggling is needed.

We also have to figure out a way to make the faces darker, as otherwise the hue we’re going to try to impose won’t look natural. Some curve correction for color is going to be necessary. As I pointed out in an earlier post, this may not be the toughest image we’ve ever seen color-wise, but it isn’t a walk in the park either, as shown by the number of people who submitted versions with obvious casts.

Is it any wonder that some of us can conquer the contrast issue but not the color, and vice versa?

Some people managed to get both at once, but I wasn’t one of them. After two disgraceful efforts, I surrendered and created one version for contrast and a second for color, then united them. I recommend that approach to others, when confronted by this difficult a challenge.

Baseball, like this image, is a very complicated game, but parts of it can be explained more simply than one might suppose. In the 1940s, Miguel González, a Cuban who had played in the American major leagues, was asked to wire a scouting report on a certain prospect in his home country. His employer expected something lengthy, but González only needed four words: “Good field, no hit.” It has now become a standard baseball expression.

The same thing can happen here, and I’ve labeled three versions as GOOD COLOR, NO SHAPE, and four others as GOOD SHAPE, NO COLOR. In doing so, I hope you will realize that these terms do not equate to BAD RESULT, but rather to HALF DOWN, HALF TO GO!

At #234 in the next post, I’ll describe how to get that second half, with a demonstration that I suggest you follow along.

This image was part of the standard curriculum for the advanced courses I taught between the late 1990s and 2011. Having taken a previous intensive course with me was a prerequisite, so these were people about as skilled as could be found in that prehistoric time. Evaluation forms at the end of each class asked the students to rate the images in terms of their instructiveness. Over more than a decade, this one consistently ranked at or near the top.

Our entrants include four from these classes. I hope you’ll agree with them about how instructive this image was. Stay tuned for the version-by-version rundown.

Dan


Dan Margulis
 



On Feb 11, 2021, at 2:19 PM, Dan Margulis via groups.io <dmargulis@...> wrote:

Because my comments on the individual versions are lengthy, I’m putting them in a separate post. This one is the introduction only.

And here are the specific comments on the Mantillas exercise.

Dan
*****************************************

201 This person created much of his facial detail through selective burning, concluding that there was too much natural variation in appearance of the three women for more than rudimentary curves correction to be helpful. The blueness of the garments seems to be a deliberate choice, partly because of concern that the mantilla at right needed a lot of attention. However, as noted in a previous post, this damaged the skintone, giving us the Norwegian appearance.

To reiterate past advice: when you think you have finished your correction, run Auto Tone. It should do nothing. If it seems to improve contrast (as it does here) even if we don’t like what it’s doing it’s a strong indication that our work is not over.

202 While #201 had a blue cast, this one’s is cyan. These things are easily measurable; see #205 for explanation.

203 Fickle fate has dictated that three of our first five entrants have chilly shadows. This one, IMHO, is quite a bit better not just than the last two but than the other half dozen we’ll see later, possibly excepting #205.  The reason? The fleshtones are more correct. Maybe this was due to some of the work this person did with H/S. It probably helped out that he has the dresses relatively dark and the skin relatively light; less chance of them affecting each other.

Interestingly, he considered the possibility that the blue dresses were objectionable, and did a second version in which the H-K script was used to neutralize them. He submitted both and told me to choose. I chose this one. The clothing is very dark without losing detail. If the skintone is of a natural hue, then bluish dresses emphasize its yellowness, a plus in an image as generally colorless as this one. As opposed to having an overall blue cast that kills the yellows.

If this guy had done some channel blending into the featureless red channel to get a bit more shape into the faces, this one would be right up there with our best.  See #205 and #220 for further comments about this version. 

204 Several versions have nice luminosity but questionable color. Here is one that does the opposite, the result of careful curving that assumed that the woman at right was wearing black but that the dresses of the other two might be reflecting some colored light. The faces seem blurrier than our best versions because the eyes and eyebrows have closed up. This may have been the result of some High Pass filtering. So I rate it GOOD COLOR, NO SHAPE.

205 Overall the impression is the same as that of #203, with bluish blacks and light fleshtones that are probably accurate for hue. #203 is slightly better, however, due to better shaping of the faces and garments, and the presence of more “irrelevant” color that might convince us that this is not a grayscale image. See #208 and #210 for other suggestions on how to improve this.

Four of our first five entries share characteristics, so let’s look at some numbers. I’ve measured LAB values for the central woman in the middle of her hair, and in the dress to the right of the cross. Results:
Central lady’s hair:
#201: 3a(10)b, purplish blue
#202: (4)a2b, green
#203: 1a6b, warm yellow
#205: 2a7b, warm yellow
Par: 5a6b, brown

Her dress:
#201: 3a(24)b, very blue, slightly purple
#202: (9)a(12)b, very cyan
#203: (2)a(16)b, very blue, slightly cyan
#205: (3)a(6)b cyan, less so than #202
Par: (1)a(3)b, a cool black

Those interested can shuffle color and luminosity layers around among these four. Not having done so, the by-the-numbers prediction is that the color of #203 and #205 would be preferred. Their numbers are questionable, which is better than the other two, which are outright wrong. Further comments about this image are in #208 and #210.

206 At first glance this one seems comparable to #203, with dark clothing and light faces. Both have good fleshtones, but this person tried to keep the dresses neutral rather than the blue of #203. Even if you think that was the right call,  I would still prefer #203 because of its better detailing in the dresses. Here the blacks have closed up so much that it’s hard to tell where the women’s hair ends and the mantilla begins. The faces came out well, however. He gave many of the individual components special attention. See #210 and #229 for suggestions on how to improve it.

207 Jeepers Creepers, where'd ya get those peepers?
Jeepers Creepers, where'd ya get those eyes?
Gosh all git up, how'd they get so lit up?
Gosh all git up, how'd they get that size?

I think from reading the description that they were modified by the program Portrait Pro.

208 This person started off on the right track with a luminosity blend of the red channel into the RGB. Thereafter he put in too much color. That might not have been so off-putting if these faces weren’t so full of noise and harsh transitions to begin with. To me it looks like these ladies have beards. A bit of softening as done by others might have made this more acceptable.

Nothing wrong with excessive color, provided you use it for blends. For example, blending 25% of this into #205 improves both versions.
 
209 From the late 1990s through 2012, I taught, 32 times, 3-day advanced courses for those who had found that the three-day basic ACT course wasn’t sufficient torture. Throughout that time, this image was part of the curriculum. The classes consistently voted it the most instructive of the 25 or so they had to deal with. And, like you, they found it quite difficult.

Our set of 41 entries include four from such classes, between ten and nineteen years old. Hopefully technique will have advanced some in that time. In evaluating them, please be advised that twenty years ago, having a dozen or so layered images of the current size open at the same time put a serious strain on the systems then available. So we worked at lower resolution, and in CMYK. I’ve converted them and upsampled them here. I’ve tried to match the cropping/positioning but it isn’t perfect, so they’re not the best choice for blending.

The first such entry is from a 2010 class in Canada. Its daddy was a police officer who specialized in forensic imaging. As that might suggest, he was pretty good at dredging out detail. I am sure he did some channel blending to get that pop in the faces. Colorwise, however, he ran into the same problem many of you did. The woman at left has blue hair, the image has enough of a cool cast to kill the yellowness in the faces, which are too pink. 

210 From a class in Italy in 2011 comes this version from our colleague Davide Barranca, who knows what Mediterranean skintone looks like. Since he also is one of those favoring very dark dresses, this effort is most directly comparable to #205 and #206. And since most of us appear to feel that Davide’s treatment of the faces is somewhat too dark, I’d say that blending 50% of it into either would improve them.

211 This attractive version addresses the question of what to do with the background. This person’s notes indicate that he realized early on that the lighting was cool, probably an overcast day. He decided to correct it to a sunnier look. He also took into account

The women are Spanish, and would normally have some facial color…I assume the wall behind the women is of a church or similar building. Hence the color could be a variety of gray towards yellowish light brown.

He did the luminosity blend of the green channel into the red, but it still lacks a little of the depth of the best versions. As for the color, I agree with him about what the background color is in real life, but I actually think we’re better off with something further away from the skin color, possibly a blue-gray. That, of course is merely a matter of taste. Further comments about this one are found in #215, #216, and #221. GOOD COLOR, NO SHAPE.

212 This person may have had different objectives than the rest of us. Roughly speaking he was after a period mood, as if was working in Photoshop 3, making it look as the film capture that it is. There’s nothing unreasonable here but I repeat the suggestion that these faces are too pale for typical Spaniards. It isn’t helped by a mild cool cast, measurable in the hair of the two women at right.

213 When I first saw this version I under-rated it. On further review it seems to me one of the best entrants. The procedure was straightforward: a luminosity blend of the green into the red, then curves and Channel Mixer to neutralize the clothing, Davide Barranca’s ALCE sharpener, and if I read the documentation correctly, a curves correction with the faces selected. That may have been what threw me at first, it seems to me that the dark areas (the eyes) are lighter than corresponding points elsewhere. And all three of the chins have gotten quite rough—softening needed. There may have been too much MMM in the faces, but on the whole, a good entrant.

214 Superficially this one is similar to #206, very dark clothing with very light faces. But at least #206 threw some color into the faces. This one is a portrait of three ice-women. The treatment suggests that the person wouldn’t like something as aggressive as #213. Still, blending 50% of it into this one would give it more life.

215 Out of the refrigerator and into the deep freeze. This person made some sophisticated blending moves and then, as I read it, had the same blueness in the clothing that plagued several people. He was wondering about how to get rid of it when the devil, who never sleeps, whispered into his ear that the solution was to blend the red channel into the blue, but fiendishly refused to specify Darken rather than Normal mode. That neutralized the clothing, all right, but it took the fleshtones along for the ride.

The color is enough to tempt one to hit the delete key. It would be a big mistake. This is just another example where the luminosity is better than the color. This is merely the most extreme example—so far, just wait till the next one. The color, instead of just being so-so, is terrible. OTOH the luminosity is among the best in the group. So, if you take the luminosity of this one and combine it with the nice color of, say, #211, the result is better than either. GOOD SHAPE, NO COLOR

216 The horror movie continues but with a similar happy ending. For the color it’s enough to quote the same numbers as in the comments to #205: for the center woman’s hair, 7a(13)b, strongly purple; and for her dress (5)a(30)b, a brilliant blue tending toward green. In short, casts far more extreme than in the group discussed there.

How to fix it? Start over. But don’t throw this version out, no way, because it has outstanding detail, maybe the best of all 40 entries. He used the green channel blending into the RGB, but especially he did some fine handwork with dodging and burning to get outstanding shape in the faces.

So, set this version aside, and make a new one where good color is the only issue. It’s quite liberating not to have to worry about channel blending or sharpening, etc., and one should be able to whip a good one up in a few minutes. Discussion of this version will continue at #234. GOOD SHAPE, NO COLOR

217 Another from my advanced classes, this one from 2002, executed by me. For twenty years old, it holds up rather well. Alas that it doesn’t have enough resolution to be a true entrant in this study.

218 For the third time in the last four images, the person has lost control of the color but left us with excellent luminosity. This image has a magenta cast as measured in several places. The most objectionable feature, however, is the variation in the faces, going from speculars, to yellowness, to mottled reds.

The excellent detailing is due, first, to the green blending, but also to a side trip into CMYK, where it’s easy to isolate and work on black objects like these dresses. As a result, he has maybe the best shape in these garments of anybody. There will be further discussion of this version at #234. GOOD SHAPE, NO COLOR.

219 A purple cast wipes out the yellowish nature of the fleshtone. The hair of the woman at right, for example, measures 7a(14)b, heavily purple. This should be checked routinely as part of the correction process.

220 Chosen for the par version. The person didn’t enclose a list of steps, so I can’t discuss technique. What I can say is that he is one of those who believes in very dark blacks, but unlike previous such versions, his faces match them appropriately. Don’t believe it? Compare it to #203 and #206, where the garments are similar but the faces much lighter.

221The color is basically correct, but the person used channel blending to neutralize all blacks. It therefore compares unfavorably to #211, with which it is superficially similar. For example, the woman at right’s hair is an attractive brown in that version and black here. Also, more agreeable color transition in the dresses.

222 Fear of a cast, as it did in #215, leaves a picture that’s too neutral. This person was aware that the ladies are Spanish and treated their skintone accordingly. But his curves must have knocked out some of the saturation in the faces as well as the blacks and the background.

His sharpening was sensitive, excluding the skin and also the background people at right. But he then desaturated the people at right. That’s the opposite of what many of us, including me, did. When I see a picture like this, it almost seems like a grayscale. Since not much can be done elsewhere, if I had to improve this version in a few seconds I’d paint saturation into the lips, the red necklace, and these people at right. This person was worried that they’d be too distracting, but I think they’re too small for that.

To prove that the basic color of this version is correct, even if too gray, put the par version on a layer above this one and set it to Saturation mode. Then, change the mode to Normal, which should reveal that this current version has somewhere along the way lost desirable detail in the eyes and eyebrows.

223 Chosen for the par version. The detailing is good, and the color is satisfactory, because by the numbers everything works. Gerald Bakker, however, hit the nail on the head

I assume that facial detail is the most important aspect of this exercise. I myself focused mainly on that. [Mine] looks remarkably similar to the par version, except for one thing: the redness of the lips. I emphasized eyes, teeth, jewelry, but forgot the mouths. 

We can go further. The original is so lacking in color that it could almost be taken for a grayscale. To prevent boredom, the experienced person finds an excuse to shoehorn some color in, the lips being an example. But it’s hard to put much extra color in the faces, the hair, the background, or the clothing. So in the par version, there is also additional color in the little boy at right and in the women’s necklaces. Coupled with slightly shapelier faces, and I would say: if you happen to like Gerald’s version, then it would be impossible not to like the par version more.

This falls in the category of those with excellent luminosity and color that—while not incorrect—can best be described as “good enough”. I’m sympathetic because I had the same problem. After two failed from-scratch efforts, I decided that I had too many balls in the air at one time, and to create one version for luminosity only and a second for color only. And I then created a third for color variation (as in the lips), used for masking in to the final product. GOOD SHAPE, NO COLOR

224 No steps were enclosed here but it looks like this person fell into the same trap as #222: the garments are extremely neutral, and he probably tried so hard to achieve it that he also neutralized the faces. Since he likes very dark dresses, it would be well to compare this one to #203 and #206 to see how important it is to get some kind of color in the women.

225 This person acknowledges having a difficult time here. The dresses get a nice middle-of-the-road treatment, but the faces are too light and too colorless.

226 Too flat, and shadows rather plugged, despite having colors that are basically correct. I once again suggest to all that when you think you’re finished, you run Image: Adjustments>Auto Tone. If you’ve done good work, it shouldn’t change things at all, let alone the major change it makes here.

227 In 2010, eight years after having done #217, I tried again in another class. In the interim, MMM had been introduced, which may account for the better color variation in the fleshtone.

228 This one is quite close to #223, for which I have a mild preference. There’s good shaping in the faces here but they are rather noisy. I also personally think that the darker clothing in this version makes the faces look weaker, even though they’re so similar to those of #223.

229 I believe the problem is that this person, instead of ending with an experimental Auto Tone, started out with it. As a result (I surmise) there are large nearly blank areas in the faces, marring an otherwise acceptable treatment.

230 We’ve seen quite a few of these before. Distinct blue cast. Using the same sample points discussed in #205, the center woman’s hair measures 3a(5)b, purple, and her dress is (3)a(17)b, very blue tending to cyan.

As with the others, the blue cast damages the inherent yellowness of the skintone and makes the women too pink.

231 Chosen for the par version. The skintone should have been softened somehow. It lacks the smoothness of, say, #223, but the variation makes this the more interesting treatment. This person was interested in getting more detail into the dresses, particularly the one at right. So he did some neat CMYK trickery with two different GCR settings. CMYK is the best way to handle detailing in black objects because they reside in the black channel where there usually isn’t much else to fork up. The usual argument against going to CMYK when the final is to be RGB is that it can eliminate certain colors that are within the RGB gamut but not the CMYK. Obviously, there aren’t any such colors in this original.

232 Chosen for the par version. By Hector Davila, this is my favorite, with a very appropriate skintone, no obvious problem, and a generally conservative treatment. A couple of interesting notes: he was (I think) the only one who wheeled out the ancient Color Balance command. Maybe it’s appropriate here, because its structure (Highlights, Midtones, Shadows) lets it more or less isolate the background, faces, and garments. Also, many of us noticed that the woman at right seems darker than the other two. Hector was (I think) the only one who tried to compensate by selecting the other two women and correcting them individually.

233 This is very comparable to #222 except that here the clothing (and hair) is darker. That would be a matter of taste. Otherwise, the two share the same problems: overly neutral skin and unsatisfactory treatment of the eyes. 

234 Paco Márquez gave a complete description of what he did here in a post on 10 February, so I won’t reiterate most of his steps, but his basic philosophy was:

Skin tones I worked on separately because here you have the woman on the right as a rosy one, the next one with a more Moorish tone and the other 2 as a mix of both. In Spain, because of the mix of races, skin tones are very varied. 

Accordingly he spent much more time than most working on the faces in isolation. Yet at the end he wasn’t wholly satisfied with the result.

This is a good thing. This is why we all go through these exercises. You submit an image you think is good, and it turns out not to be so. You don’t need me to say it outright because it’s obvious by comparing yours to a better version.

It’s not enough, though, to just say that the other is better because it’s easy to pick the wrong reason. In this study, many people think they have inferior versions because their skintone is too pink. Sometimes that is the correct answer in the sense that the rest of the image is OK. More frequently, they’re mistaking the symptom for the disease. The problem is that the version has a blue cast, which kills the yellow component of the flesh. 

Similarly, Paco’s self-evaluation:

I missed the subdued colors in the mantillas which show up well in the par version, and screwed up by desaturating them too much. Apart from that I feel I pulled out a credible correction of the original. 

I think that this again mistakes symptom for disease, and hereby offer a demonstration that I recommend to everyone, because it says a lot about how to handle this image. Here we go:

Gather the ingredients:
2 copies of the par, #241.
1 copy of Paco’s #234.
1 copy of the horrendously blue #216.
1 copy of #218, with its swarthy skintone.

FIRST TEST.
1) Add a duplicate layer to #234.
2) Set it to Color mode. Appearance doesn’t change because the two layers are identical.
3) Shift-drag this Color layer onto one copy of the par.

Wow! It’s very unusual to find an image with better color than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. Paco’s hard work paid off. The par is now improved by the addition of his color.

The catch: Paco didn’t say anything that indicated he knew how truly outstanding his color was. And if he underestimated how good is color was, it’s safe to guess that he also underestimated how bad his luminosity was. Therefore,

4) Change the Color layer’s mode to Normal, and toggle back and forth between the two modes.

Disaster! The drawback Paco pointed out was only one of many. Best approach at this point: start over, and make a new version, worrying only about contrast. It may make it easier to actually produce a grayscale. And why not? The color problem is already solved.

SECOND TEST.
1) Add a duplicate layer to #216.
2) Set it to Luminosity mode. Appearance doesn’t change because the two layers are identical.
3) Shift-drag this Luminosity layer onto the other copy of the par.

Wow! It’s very unusual to find an image with better luminosity than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. 

4) Change the Luminosity layer’s mode to Normal, and toggle back and forth between the two modes.

Disaster!

We know that two wrongs do not make a right, so there is no point of blending the color of #16 with Paco's luminosity, unless we are trying to make mockups for a horror movie. However, the question is whether two rights might make a wrong. The logical prediction would be that the luminosity of #216 and the color of #234 would make a dynamite pairing, but...

THIRD TEST.
1) Shift-drag the Luminosity layer from the second test onto #234.

Not bad, but perhaps not as good as expected. The shape of the faces is great, but the color isn't what Paco had in mind IMHO. His color depends on faces darker than these. I'm not saying that the faces of #218 are too light in isolation, just that they're too light in combination with Paco's color choice. Similarly, the clothing of #218 is much lighter than in #234. So,

2) Reduce the opacity of the Luminosity layer to 50%.

I see this as an improvement over #234 as submitted. But, we must concede that it saves 50% of Paco's original luminosity, which is 50% more than deserves saving. Therefore,

FOURTH TEST.
1) Restore the Luminosity layer to 100% opacity.
2) To it, apply #218, (replacing #216).

I don't think #218 is quite at the level of #216 for quality of luminosity, but it's pretty good, and is much more in line with Paco's views on how dark to make the faces and clothing. Therefore, 100% opacity is just fine.

3) Get rid of the noise in the faces that #218 brought along for the ride.

And there you have it. I find this combined version to be preferable to the par, even though neither #218 nor #234 as originally submitted is worthy of much notice. This is why I wound up doing my own version this way, except that it did not come out as well. Let us remember, therefore, that there is no disgrace in evaluating a certain version as GOOD COLOR, NO SHAPE.

235 Similar to #230, except bluer and flatter. Once again: run Auto Tone at the end. The failure to do so accounts for much of the flatness here.

236 This person tried a lot of things before becoming frustrated. This resulted in a hybrid of issues. The dresses morph into a deep blue that is disagreeable. Yet it isn’t a true blue cast, because the skintone isn’t affected much. It’s too light, and needs some noise removed, but that’s another story.  Here I’ll just describe the blend that fixes odd blue areas.

On a duplicate layer, apply the blue channel to the RGB in Lighten mode. This works because the blue is commonly darker than both others, except in blue objects, and notably in lipstick, where the green is usually the darkest. And a few other exceptions that are not found here.

This move completely neutralizes all blues. I would think that’s inappropriate for the boy at right, so I’d add a layer mask and paint his color back in. Also, the blend completely neutralizes the blue areas in the dress, a small amount of which might be desirable. So reduce layer opacity to taste. And then the picture is more respectable.

237 The image has a mild blue cast, not as bad as some of the others, but still presumably is partly responsible for the Scandinavian-style skin. Here the skin has been softened to eliminate the noise, but shape was lost in the process. To get an idea of how it might be improved, try blending in #216, in Darken mode to preserve the blacks of this version. Setting the blend layer at 50% Luminosity should get a decent result.

238 Chosen for the par version. It’s mine, and if you read the comments on #234 you know what I did.

239 This one is quite similar to the par version—except for the overly light, overly pink faces. We’ve seen several version where the fatal pinkness was caused by an overall blue cast, but that isn’t the case here. I’m not sure how it came about.  Everywhere I measure in the skin the A value comes out significantly more positive than the B. With Mediterranean types the reverse is usually true, except where they’re wearing makeup.

Whatever put the pinkness into the faces also put it into the whites of the eyes. I measure those of the center woman as nearly 20a. It’s true that these areas should be somewhat pink because of the blood vessels in the eye, but they shouldn’t be as bad as this. Making them more neutral makes the surrounding skin more palatable.

240 The original gave us a lot to work with in the clothing but pitifully little in the faces. Accordingly, our failures were almost always due to inadequate treatment of them, or issues of color that did bad things to them. As best I recall this is the only version with a big problem in the detailing of the clothing. Everything is closed up. To get an idea of what’s missing, grab a copy of #231, where the guy went to a lot of trouble to get detailing in the blacks. Blend it into this version, 50%, Lighten mode. That won’t affect the faces, but it will show how much better this version would look with more definition in the blacks.

241 The par, blending the versions designated above, with each given 20% weight.


KENT SOUTHERS
 


Dan,

Thanks for the feedback.

Upon review, I noted that the blacks in the two left dresses ranged from 30's - 60's.  For me, that's about where I wanted the blacks to be.

As I mentioned in my writeup, I noted the dress on the right being of a different material (velvet ?), with a different reflectance level, so I accepted the darker values there.
After reading your comments, I pulled up some of the values in the dress in the right, but opacity masked the other areas in the left two dresses to hold them from going higher than the 60's.  Meanwhile pulling the right dress into the 50's - 60's (and still some darker areas)

As to means of correction, I opted to NOT use another image as an overlay (since, I don't have that option in the real world), but did heed the point about the crushed blacks ... mostly in the velvet.  Which begs the question of luminance for blacks ... what values do you expect black to be for a given sheen?  (Bearing in mind the original file had values around 2 in the deep velvet).

Thanks,


Kent Southers, CMRP
southers3@...



From: colortheory@groups.io <colortheory@groups.io> on behalf of Dan Margulis via groups.io <dmargulis@...>
Sent: Thursday, February 11, 2021 9:52 PM
To: colortheory@groups.io <colortheory@groups.io>
Subject: Re: [colortheory] Mantillas: Dan's comments
 


On Feb 11, 2021, at 2:19 PM, Dan Margulis via groups.io <dmargulis@...> wrote:

Because my comments on the individual versions are lengthy, I’m putting them in a separate post. This one is the introduction only.

And here are the specific comments on the Mantillas exercise.

Dan
*****************************************

201 This person created much of his facial detail through selective burning, concluding that there was too much natural variation in appearance of the three women for more than rudimentary curves correction to be helpful. The blueness of the garments seems to be a deliberate choice, partly because of concern that the mantilla at right needed a lot of attention. However, as noted in a previous post, this damaged the skintone, giving us the Norwegian appearance.

To reiterate past advice: when you think you have finished your correction, run Auto Tone. It should do nothing. If it seems to improve contrast (as it does here) even if we don’t like what it’s doing it’s a strong indication that our work is not over.

202 While #201 had a blue cast, this one’s is cyan. These things are easily measurable; see #205 for explanation.

203 Fickle fate has dictated that three of our first five entrants have chilly shadows. This one, IMHO, is quite a bit better not just than the last two but than the other half dozen we’ll see later, possibly excepting #205.  The reason? The fleshtones are more correct. Maybe this was due to some of the work this person did with H/S. It probably helped out that he has the dresses relatively dark and the skin relatively light; less chance of them affecting each other.

Interestingly, he considered the possibility that the blue dresses were objectionable, and did a second version in which the H-K script was used to neutralize them. He submitted both and told me to choose. I chose this one. The clothing is very dark without losing detail. If the skintone is of a natural hue, then bluish dresses emphasize its yellowness, a plus in an image as generally colorless as this one. As opposed to having an overall blue cast that kills the yellows.

If this guy had done some channel blending into the featureless red channel to get a bit more shape into the faces, this one would be right up there with our best.  See #205 and #220 for further comments about this version. 

204 Several versions have nice luminosity but questionable color. Here is one that does the opposite, the result of careful curving that assumed that the woman at right was wearing black but that the dresses of the other two might be reflecting some colored light. The faces seem blurrier than our best versions because the eyes and eyebrows have closed up. This may have been the result of some High Pass filtering. So I rate it GOOD COLOR, NO SHAPE.

205 Overall the impression is the same as that of #203, with bluish blacks and light fleshtones that are probably accurate for hue. #203 is slightly better, however, due to better shaping of the faces and garments, and the presence of more “irrelevant” color that might convince us that this is not a grayscale image. See #208 and #210 for other suggestions on how to improve this.

Four of our first five entries share characteristics, so let’s look at some numbers. I’ve measured LAB values for the central woman in the middle of her hair, and in the dress to the right of the cross. Results:
Central lady’s hair:
#201: 3a(10)b, purplish blue
#202: (4)a2b, green
#203: 1a6b, warm yellow
#205: 2a7b, warm yellow
Par: 5a6b, brown

Her dress:
#201: 3a(24)b, very blue, slightly purple
#202: (9)a(12)b, very cyan
#203: (2)a(16)b, very blue, slightly cyan
#205: (3)a(6)b cyan, less so than #202
Par: (1)a(3)b, a cool black

Those interested can shuffle color and luminosity layers around among these four. Not having done so, the by-the-numbers prediction is that the color of #203 and #205 would be preferred. Their numbers are questionable, which is better than the other two, which are outright wrong. Further comments about this image are in #208 and #210.

206 At first glance this one seems comparable to #203, with dark clothing and light faces. Both have good fleshtones, but this person tried to keep the dresses neutral rather than the blue of #203. Even if you think that was the right call,  I would still prefer #203 because of its better detailing in the dresses. Here the blacks have closed up so much that it’s hard to tell where the women’s hair ends and the mantilla begins. The faces came out well, however. He gave many of the individual components special attention. See #210 and #229 for suggestions on how to improve it.

207 Jeepers Creepers, where'd ya get those peepers?
Jeepers Creepers, where'd ya get those eyes?
Gosh all git up, how'd they get so lit up?
Gosh all git up, how'd they get that size?

I think from reading the description that they were modified by the program Portrait Pro.

208 This person started off on the right track with a luminosity blend of the red channel into the RGB. Thereafter he put in too much color. That might not have been so off-putting if these faces weren’t so full of noise and harsh transitions to begin with. To me it looks like these ladies have beards. A bit of softening as done by others might have made this more acceptable.

Nothing wrong with excessive color, provided you use it for blends. For example, blending 25% of this into #205 improves both versions.
 
209 From the late 1990s through 2012, I taught, 32 times, 3-day advanced courses for those who had found that the three-day basic ACT course wasn’t sufficient torture. Throughout that time, this image was part of the curriculum. The classes consistently voted it the most instructive of the 25 or so they had to deal with. And, like you, they found it quite difficult.

Our set of 41 entries include four from such classes, between ten and nineteen years old. Hopefully technique will have advanced some in that time. In evaluating them, please be advised that twenty years ago, having a dozen or so layered images of the current size open at the same time put a serious strain on the systems then available. So we worked at lower resolution, and in CMYK. I’ve converted them and upsampled them here. I’ve tried to match the cropping/positioning but it isn’t perfect, so they’re not the best choice for blending.

The first such entry is from a 2010 class in Canada. Its daddy was a police officer who specialized in forensic imaging. As that might suggest, he was pretty good at dredging out detail. I am sure he did some channel blending to get that pop in the faces. Colorwise, however, he ran into the same problem many of you did. The woman at left has blue hair, the image has enough of a cool cast to kill the yellowness in the faces, which are too pink. 

210 From a class in Italy in 2011 comes this version from our colleague Davide Barranca, who knows what Mediterranean skintone looks like. Since he also is one of those favoring very dark dresses, this effort is most directly comparable to #205 and #206. And since most of us appear to feel that Davide’s treatment of the faces is somewhat too dark, I’d say that blending 50% of it into either would improve them.

211 This attractive version addresses the question of what to do with the background. This person’s notes indicate that he realized early on that the lighting was cool, probably an overcast day. He decided to correct it to a sunnier look. He also took into account

The women are Spanish, and would normally have some facial color…I assume the wall behind the women is of a church or similar building. Hence the color could be a variety of gray towards yellowish light brown.

He did the luminosity blend of the green channel into the red, but it still lacks a little of the depth of the best versions. As for the color, I agree with him about what the background color is in real life, but I actually think we’re better off with something further away from the skin color, possibly a blue-gray. That, of course is merely a matter of taste. Further comments about this one are found in #215, #216, and #221. GOOD COLOR, NO SHAPE.

212 This person may have had different objectives than the rest of us. Roughly speaking he was after a period mood, as if was working in Photoshop 3, making it look as the film capture that it is. There’s nothing unreasonable here but I repeat the suggestion that these faces are too pale for typical Spaniards. It isn’t helped by a mild cool cast, measurable in the hair of the two women at right.

213 When I first saw this version I under-rated it. On further review it seems to me one of the best entrants. The procedure was straightforward: a luminosity blend of the green into the red, then curves and Channel Mixer to neutralize the clothing, Davide Barranca’s ALCE sharpener, and if I read the documentation correctly, a curves correction with the faces selected. That may have been what threw me at first, it seems to me that the dark areas (the eyes) are lighter than corresponding points elsewhere. And all three of the chins have gotten quite rough—softening needed. There may have been too much MMM in the faces, but on the whole, a good entrant.

214 Superficially this one is similar to #206, very dark clothing with very light faces. But at least #206 threw some color into the faces. This one is a portrait of three ice-women. The treatment suggests that the person wouldn’t like something as aggressive as #213. Still, blending 50% of it into this one would give it more life.

215 Out of the refrigerator and into the deep freeze. This person made some sophisticated blending moves and then, as I read it, had the same blueness in the clothing that plagued several people. He was wondering about how to get rid of it when the devil, who never sleeps, whispered into his ear that the solution was to blend the red channel into the blue, but fiendishly refused to specify Darken rather than Normal mode. That neutralized the clothing, all right, but it took the fleshtones along for the ride.

The color is enough to tempt one to hit the delete key. It would be a big mistake. This is just another example where the luminosity is better than the color. This is merely the most extreme example—so far, just wait till the next one. The color, instead of just being so-so, is terrible. OTOH the luminosity is among the best in the group. So, if you take the luminosity of this one and combine it with the nice color of, say, #211, the result is better than either. GOOD SHAPE, NO COLOR

216 The horror movie continues but with a similar happy ending. For the color it’s enough to quote the same numbers as in the comments to #205: for the center woman’s hair, 7a(13)b, strongly purple; and for her dress (5)a(30)b, a brilliant blue tending toward green. In short, casts far more extreme than in the group discussed there.

How to fix it? Start over. But don’t throw this version out, no way, because it has outstanding detail, maybe the best of all 40 entries. He used the green channel blending into the RGB, but especially he did some fine handwork with dodging and burning to get outstanding shape in the faces.

So, set this version aside, and make a new one where good color is the only issue. It’s quite liberating not to have to worry about channel blending or sharpening, etc., and one should be able to whip a good one up in a few minutes. Discussion of this version will continue at #234. GOOD SHAPE, NO COLOR

217 Another from my advanced classes, this one from 2002, executed by me. For twenty years old, it holds up rather well. Alas that it doesn’t have enough resolution to be a true entrant in this study.

218 For the third time in the last four images, the person has lost control of the color but left us with excellent luminosity. This image has a magenta cast as measured in several places. The most objectionable feature, however, is the variation in the faces, going from speculars, to yellowness, to mottled reds.

The excellent detailing is due, first, to the green blending, but also to a side trip into CMYK, where it’s easy to isolate and work on black objects like these dresses. As a result, he has maybe the best shape in these garments of anybody. There will be further discussion of this version at #234. GOOD SHAPE, NO COLOR.

219 A purple cast wipes out the yellowish nature of the fleshtone. The hair of the woman at right, for example, measures 7a(14)b, heavily purple. This should be checked routinely as part of the correction process.

220 Chosen for the par version. The person didn’t enclose a list of steps, so I can’t discuss technique. What I can say is that he is one of those who believes in very dark blacks, but unlike previous such versions, his faces match them appropriately. Don’t believe it? Compare it to #203 and #206, where the garments are similar but the faces much lighter.

221The color is basically correct, but the person used channel blending to neutralize all blacks. It therefore compares unfavorably to #211, with which it is superficially similar. For example, the woman at right’s hair is an attractive brown in that version and black here. Also, more agreeable color transition in the dresses.

222 Fear of a cast, as it did in #215, leaves a picture that’s too neutral. This person was aware that the ladies are Spanish and treated their skintone accordingly. But his curves must have knocked out some of the saturation in the faces as well as the blacks and the background.

His sharpening was sensitive, excluding the skin and also the background people at right. But he then desaturated the people at right. That’s the opposite of what many of us, including me, did. When I see a picture like this, it almost seems like a grayscale. Since not much can be done elsewhere, if I had to improve this version in a few seconds I’d paint saturation into the lips, the red necklace, and these people at right. This person was worried that they’d be too distracting, but I think they’re too small for that.

To prove that the basic color of this version is correct, even if too gray, put the par version on a layer above this one and set it to Saturation mode. Then, change the mode to Normal, which should reveal that this current version has somewhere along the way lost desirable detail in the eyes and eyebrows.

223 Chosen for the par version. The detailing is good, and the color is satisfactory, because by the numbers everything works. Gerald Bakker, however, hit the nail on the head

I assume that facial detail is the most important aspect of this exercise. I myself focused mainly on that. [Mine] looks remarkably similar to the par version, except for one thing: the redness of the lips. I emphasized eyes, teeth, jewelry, but forgot the mouths. 

We can go further. The original is so lacking in color that it could almost be taken for a grayscale. To prevent boredom, the experienced person finds an excuse to shoehorn some color in, the lips being an example. But it’s hard to put much extra color in the faces, the hair, the background, or the clothing. So in the par version, there is also additional color in the little boy at right and in the women’s necklaces. Coupled with slightly shapelier faces, and I would say: if you happen to like Gerald’s version, then it would be impossible not to like the par version more.

This falls in the category of those with excellent luminosity and color that—while not incorrect—can best be described as “good enough”. I’m sympathetic because I had the same problem. After two failed from-scratch efforts, I decided that I had too many balls in the air at one time, and to create one version for luminosity only and a second for color only. And I then created a third for color variation (as in the lips), used for masking in to the final product. GOOD SHAPE, NO COLOR

224 No steps were enclosed here but it looks like this person fell into the same trap as #222: the garments are extremely neutral, and he probably tried so hard to achieve it that he also neutralized the faces. Since he likes very dark dresses, it would be well to compare this one to #203 and #206 to see how important it is to get some kind of color in the women.

225 This person acknowledges having a difficult time here. The dresses get a nice middle-of-the-road treatment, but the faces are too light and too colorless.

226 Too flat, and shadows rather plugged, despite having colors that are basically correct. I once again suggest to all that when you think you’re finished, you run Image: Adjustments>Auto Tone. If you’ve done good work, it shouldn’t change things at all, let alone the major change it makes here.

227 In 2010, eight years after having done #217, I tried again in another class. In the interim, MMM had been introduced, which may account for the better color variation in the fleshtone.

228 This one is quite close to #223, for which I have a mild preference. There’s good shaping in the faces here but they are rather noisy. I also personally think that the darker clothing in this version makes the faces look weaker, even though they’re so similar to those of #223.

229 I believe the problem is that this person, instead of ending with an experimental Auto Tone, started out with it. As a result (I surmise) there are large nearly blank areas in the faces, marring an otherwise acceptable treatment.

230 We’ve seen quite a few of these before. Distinct blue cast. Using the same sample points discussed in #205, the center woman’s hair measures 3a(5)b, purple, and her dress is (3)a(17)b, very blue tending to cyan.

As with the others, the blue cast damages the inherent yellowness of the skintone and makes the women too pink.

231 Chosen for the par version. The skintone should have been softened somehow. It lacks the smoothness of, say, #223, but the variation makes this the more interesting treatment. This person was interested in getting more detail into the dresses, particularly the one at right. So he did some neat CMYK trickery with two different GCR settings. CMYK is the best way to handle detailing in black objects because they reside in the black channel where there usually isn’t much else to fork up. The usual argument against going to CMYK when the final is to be RGB is that it can eliminate certain colors that are within the RGB gamut but not the CMYK. Obviously, there aren’t any such colors in this original.

232 Chosen for the par version. By Hector Davila, this is my favorite, with a very appropriate skintone, no obvious problem, and a generally conservative treatment. A couple of interesting notes: he was (I think) the only one who wheeled out the ancient Color Balance command. Maybe it’s appropriate here, because its structure (Highlights, Midtones, Shadows) lets it more or less isolate the background, faces, and garments. Also, many of us noticed that the woman at right seems darker than the other two. Hector was (I think) the only one who tried to compensate by selecting the other two women and correcting them individually.

233 This is very comparable to #222 except that here the clothing (and hair) is darker. That would be a matter of taste. Otherwise, the two share the same problems: overly neutral skin and unsatisfactory treatment of the eyes. 

234 Paco Márquez gave a complete description of what he did here in a post on 10 February, so I won’t reiterate most of his steps, but his basic philosophy was:

Skin tones I worked on separately because here you have the woman on the right as a rosy one, the next one with a more Moorish tone and the other 2 as a mix of both. In Spain, because of the mix of races, skin tones are very varied. 

Accordingly he spent much more time than most working on the faces in isolation. Yet at the end he wasn’t wholly satisfied with the result.

This is a good thing. This is why we all go through these exercises. You submit an image you think is good, and it turns out not to be so. You don’t need me to say it outright because it’s obvious by comparing yours to a better version.

It’s not enough, though, to just say that the other is better because it’s easy to pick the wrong reason. In this study, many people think they have inferior versions because their skintone is too pink. Sometimes that is the correct answer in the sense that the rest of the image is OK. More frequently, they’re mistaking the symptom for the disease. The problem is that the version has a blue cast, which kills the yellow component of the flesh. 

Similarly, Paco’s self-evaluation:

I missed the subdued colors in the mantillas which show up well in the par version, and screwed up by desaturating them too much. Apart from that I feel I pulled out a credible correction of the original. 

I think that this again mistakes symptom for disease, and hereby offer a demonstration that I recommend to everyone, because it says a lot about how to handle this image. Here we go:

Gather the ingredients:
2 copies of the par, #241.
1 copy of Paco’s #234.
1 copy of the horrendously blue #216.
1 copy of #218, with its swarthy skintone.

FIRST TEST.
1) Add a duplicate layer to #234.
2) Set it to Color mode. Appearance doesn’t change because the two layers are identical.
3) Shift-drag this Color layer onto one copy of the par.

Wow! It’s very unusual to find an image with better color than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. Paco’s hard work paid off. The par is now improved by the addition of his color.

The catch: Paco didn’t say anything that indicated he knew how truly outstanding his color was. And if he underestimated how good is color was, it’s safe to guess that he also underestimated how bad his luminosity was. Therefore,

4) Change the Color layer’s mode to Normal, and toggle back and forth between the two modes.

Disaster! The drawback Paco pointed out was only one of many. Best approach at this point: start over, and make a new version, worrying only about contrast. It may make it easier to actually produce a grayscale. And why not? The color problem is already solved.

SECOND TEST.
1) Add a duplicate layer to #216.
2) Set it to Luminosity mode. Appearance doesn’t change because the two layers are identical.
3) Shift-drag this Luminosity layer onto the other copy of the par.

Wow! It’s very unusual to find an image with better luminosity than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. 

4) Change the Luminosity layer’s mode to Normal, and toggle back and forth between the two modes.

Disaster!

We know that two wrongs do not make a right, so there is no point of blending the color of #16 with Paco's luminosity, unless we are trying to make mockups for a horror movie. However, the question is whether two rights might make a wrong. The logical prediction would be that the luminosity of #216 and the color of #234 would make a dynamite pairing, but...

THIRD TEST.
1) Shift-drag the Luminosity layer from the second test onto #234.

Not bad, but perhaps not as good as expected. The shape of the faces is great, but the color isn't what Paco had in mind IMHO. His color depends on faces darker than these. I'm not saying that the faces of #218 are too light in isolation, just that they're too light in combination with Paco's color choice. Similarly, the clothing of #218 is much lighter than in #234. So,

2) Reduce the opacity of the Luminosity layer to 50%.

I see this as an improvement over #234 as submitted. But, we must concede that it saves 50% of Paco's original luminosity, which is 50% more than deserves saving. Therefore,

FOURTH TEST.
1) Restore the Luminosity layer to 100% opacity.
2) To it, apply #218, (replacing #216).

I don't think #218 is quite at the level of #216 for quality of luminosity, but it's pretty good, and is much more in line with Paco's views on how dark to make the faces and clothing. Therefore, 100% opacity is just fine.

3) Get rid of the noise in the faces that #218 brought along for the ride.

And there you have it. I find this combined version to be preferable to the par, even though neither #218 nor #234 as originally submitted is worthy of much notice. This is why I wound up doing my own version this way, except that it did not come out as well. Let us remember, therefore, that there is no disgrace in evaluating a certain version as GOOD COLOR, NO SHAPE.

235 Similar to #230, except bluer and flatter. Once again: run Auto Tone at the end. The failure to do so accounts for much of the flatness here.

236 This person tried a lot of things before becoming frustrated. This resulted in a hybrid of issues. The dresses morph into a deep blue that is disagreeable. Yet it isn’t a true blue cast, because the skintone isn’t affected much. It’s too light, and needs some noise removed, but that’s another story.  Here I’ll just describe the blend that fixes odd blue areas.

On a duplicate layer, apply the blue channel to the RGB in Lighten mode. This works because the blue is commonly darker than both others, except in blue objects, and notably in lipstick, where the green is usually the darkest. And a few other exceptions that are not found here.

This move completely neutralizes all blues. I would think that’s inappropriate for the boy at right, so I’d add a layer mask and paint his color back in. Also, the blend completely neutralizes the blue areas in the dress, a small amount of which might be desirable. So reduce layer opacity to taste. And then the picture is more respectable.

237 The image has a mild blue cast, not as bad as some of the others, but still presumably is partly responsible for the Scandinavian-style skin. Here the skin has been softened to eliminate the noise, but shape was lost in the process. To get an idea of how it might be improved, try blending in #216, in Darken mode to preserve the blacks of this version. Setting the blend layer at 50% Luminosity should get a decent result.

238 Chosen for the par version. It’s mine, and if you read the comments on #234 you know what I did.

239 This one is quite similar to the par version—except for the overly light, overly pink faces. We’ve seen several version where the fatal pinkness was caused by an overall blue cast, but that isn’t the case here. I’m not sure how it came about.  Everywhere I measure in the skin the A value comes out significantly more positive than the B. With Mediterranean types the reverse is usually true, except where they’re wearing makeup.

Whatever put the pinkness into the faces also put it into the whites of the eyes. I measure those of the center woman as nearly 20a. It’s true that these areas should be somewhat pink because of the blood vessels in the eye, but they shouldn’t be as bad as this. Making them more neutral makes the surrounding skin more palatable.

240 The original gave us a lot to work with in the clothing but pitifully little in the faces. Accordingly, our failures were almost always due to inadequate treatment of them, or issues of color that did bad things to them. As best I recall this is the only version with a big problem in the detailing of the clothing. Everything is closed up. To get an idea of what’s missing, grab a copy of #231, where the guy went to a lot of trouble to get detailing in the blacks. Blend it into this version, 50%, Lighten mode. That won’t affect the faces, but it will show how much better this version would look with more definition in the blacks.

241 The par, blending the versions designated above, with each given 20% weight.


David Remington
 
Edited

Dan,

Thanks again for the generous and detailed analysis. Mine is version 214 the visitation from the North. I agree with your feedback that the faces are too light and have insufficient modeling and the clothing is a bit too dark. You are also correct that I find some of the other version overdone and not accurate or realistic to my eye. More illustration than photograph. There is a medium in between though. Version 220 comes closer to my taste than 213. This type of work that is really retouching and colorizing is very different from my everyday work. It will be interesting to see how my approach fits with the other images you have queued up. I will experiment with the channel blending.

Thanks again.

David


Gerald Bakker
 

On Fri, Feb 12, 2021 at 04:52 AM, Dan Margulis wrote:
223 Chosen for the par version. The detailing is good, and the color is satisfactory, because by the numbers everything works. Gerald Bakker, however, hit the nail on the head
 
 
I assume that facial detail is the most important aspect of this exercise. I myself focused mainly on that. [Mine] looks remarkably similar to the par version, except for one thing: the redness of the lips. I emphasized eyes, teeth, jewelry, but forgot the mouths. 
 
We can go further. The original is so lacking in color that it could almost be taken for a grayscale. To prevent boredom, the experienced person finds an excuse to shoehorn some color in, the lips being an example. But it’s hard to put much extra color in the faces, the hair, the background, or the clothing. So in the par version, there is also additional color in the little boy at right and in the women’s necklaces. Coupled with slightly shapelier faces, and I would say: if you happen to like Gerald’s version, then it would be impossible not to like the par version more.
I am impressed by Dan's very extensive comments, but somewhat puzzled by this paragraph. So we strive to make the image look more colorful by giving some small or less important elements extra color (because the main elements cannot stand much more color). That makes sense, but if the choice is between the people on the very right and the background wall, I am not so sure. In fact, I deliberately desaturated the boy and woman on the right because I thought their colorful clothes could be a distraction. Instead, my background wall has more color than the par version's, and to me that works better than a more colorful boy.
I certainly agree about the lips and the red necklace though.  
--
Gerald Bakker
https://geraldbakker.nl


KENT SOUTHERS
 

I should clarify ... when I mentioned values in the 50-60's, etc. , that was on the RGB scale.  L values correspond mostly in the 20's with shadow areas of the blacks reaching into low teens, highlight areas reaching in low 30's.

Kent Southers, CMRP
southers3@...



From: colortheory@groups.io <colortheory@groups.io> on behalf of Gerald Bakker <gc.bakker@...>
Sent: Friday, February 12, 2021 9:48 AM
To: colortheory@groups.io <colortheory@groups.io>
Subject: Re: [colortheory] Mantillas: Dan's comments
 
On Fri, Feb 12, 2021 at 04:52 AM, Dan Margulis wrote:
223 Chosen for the par version. The detailing is good, and the color is satisfactory, because by the numbers everything works. Gerald Bakker, however, hit the nail on the head
 
 
I assume that facial detail is the most important aspect of this exercise. I myself focused mainly on that. [Mine] looks remarkably similar to the par version, except for one thing: the redness of the lips. I emphasized eyes, teeth, jewelry, but forgot the mouths. 
 
We can go further. The original is so lacking in color that it could almost be taken for a grayscale. To prevent boredom, the experienced person finds an excuse to shoehorn some color in, the lips being an example. But it’s hard to put much extra color in the faces, the hair, the background, or the clothing. So in the par version, there is also additional color in the little boy at right and in the women’s necklaces. Coupled with slightly shapelier faces, and I would say: if you happen to like Gerald’s version, then it would be impossible not to like the par version more.
I am impressed by Dan's very extensive comments, but somewhat puzzled by this paragraph. So we strive to make the image look more colorful by giving some small or less important elements extra color (because the main elements cannot stand much more color). That makes sense, but if the choice is between the people on the very right and the background wall, I am not so sure. In fact, I deliberately desaturated the boy and woman on the right because I thought their colorful clothes could be a distraction. Instead, my background wall has more color than the par version's, and to me that works better than a more colorful boy.
I certainly agree about the lips and the red necklace though.  
--
Gerald Bakker
https://geraldbakker.nl


Paco
 

On 234, went through all the Tests and I can see how much, for the better, these merges make. Now that we know the how. Please explain the why.

Paco


Bill Theis
 
Edited

Thanks Dan for this very enlightening and useful exercise (which somehow I think I failed on once before)

mine was 218 which Dan judged as good shape lousy color

Guilty.  "Swarthy" and noisy but doing the 4 Test exercise which effectively blends my luminosity with 234's color gives me hope.  So I looked at my failings in color: 

Since I didn't know the skin color of the Mantillas, I actually looked at the caucasian woman to the right of them to set reasonable (to me) skin color at a reddish (a,b)=(35,20) since when I made b>a for this face the MANTILLAS faces became obviously badly yellow.  So I though it must be a cold day maybe, make the caucasian color more red so that the resulting Mantillas skin tone become about (20,25).  234 which had the best color had this at (19,15), considerably less yellow.  So back to color by the numbers:  "Positive A and B, B usually higher.  Caucasian have higher A values.. as a rule a higher A should make you suspicious that the image is too purple."  Why is it preferable here?  I do see that it is better with my eyes but I have so much more trust in my eyedropper.  (Eyes are getting old)

I did put a curve to reproduce the colors of 234 and although this was obviously much better, I didn't know to do this at the time since I pretty blindly follow color by the numbers.  Which brings up the same problem I had with the Lobby with the greenery which should be "if B is less than 1.5 the (absolute value) of A suggests the image is too blue".  The par version for the Lobby had B LESS THAN (absolute value) of A on the order of (-39,20) in the top palm, something pointed out by Gerald Bakker.  Any answer for this?  (I checked the forums and no one has answered this)

Dan knows my strong suite is not skin, but I am confused since I went by the book.

Bill Theis


Robert S Baldassano
 

Dan thanks for the inputs. I did as you suggested and sure enough the blue cast is gone. Here is the update. I also tried blurring the A & B channels after this version but did not like how it turned out. But this is definitely better than what I had submitted.

 

 

 

Sent from Mail for Windows 10

 


Dan Margulis
 



On Feb 12, 2021, at 2:16 PM, williamtheis via groups.io <williamtheis@...> wrote:

Since I didn't know the skin color of the Mantillas, I actually looked at the caucasian woman to the right of them to set reasonable (to me) skin color at a reddish (a,b)=(35,20) since when I made b>a for this face the MANTILLAS faces became obviously badly yellow.  So I though it must be a cold day maybe, make the caucasian color more red so that the resulting Mantillas skin tone become about (20,25).  234 which had the best color had this at (19,15), considerably less yellow.  So back to color by the numbers:  "Positive A and B, B usually higher.  Caucasian have higher A values.. as a rule a higher A should make you suspicious that the image is too purple."  Why is it preferable here?  I do see that it is better with my eyes but I have so much more trust in my eyedropper.  (Eyes are getting old)

AB values for skin have to be taken with several grains of salt because of how they interrelate with the L. For serious color by the numbers I suggest changing the Info palette to read CMYK values, which will give a better indication of how saturated the skin is.

That said, my LAB readings are different than yours. I got them by selecting simultaneously four areas in the women’s faces, trying to avoid areas that were likely to be affected by makeup. Then I did Filter: Blur>Average and measured the result in several of the better versions, getting the following results in some of the better versions:

#218 (Bill) 74L 25a 26b
#223 (Gerald) 82L 16a 22b
#232 (Hector) 78L 13a 24b
#234 (Paco) 72L 21a 19b
#238 (me) 76L 15a 18b
#241 (par) 77L 15a 18b

That’s a wide range, with Paco and Hector being the extremes of redness and yellowness. The rest of us prefer a hue somewhere in the middle, but I find both acceptable.

The problem with your skintone, therefore, has nothing to do with the AB ratio and everything to do with it being significantly more saturated than the others. A lot of people had Swedish skintone, way too pale. This is the other extreme, too beet-red.

Dan


Doug Schafer
 

Dan, Thanks for such a lengthy review and providing so much info. I will read again and try some of your ideas to see the combinations via blending 2 images.

I do have 1 question. You did not mention it at all that I read; so I'm assuming is didn't seem important to you...but it sure bugged me...Mine was 206.

Mantillas are veils with semi transparent areas we see thru to the background. Yet only a few of us corrected the veils to eliminate the unrealistic (to me) colors in the top parts of mantillas in majority of the images.
So the question: is the better treatment to fix the transparency colors to match the background or accept the bright mis-matched colors....which I agree if they were colorful decorations on the veils would look nice.  But I think the colors are in error???
Can you please comment so we can better appreciate what to do in future veil images; often associated with brides which would seem to need corrections to avoid bright saturated colors in the veils not seen thru from the background.

Doug Schafer


Dan Margulis
 



On Feb 12, 2021, at 2:51 AM, KENT SOUTHERS <southers3@...> wrote:

As to means of correction, I opted to NOT use another image as an overlay (since, I don't have that option in the real world), but did heed the point about the crushed blacks ... mostly in the velvet.  Which begs the question of luminance for blacks ... what values do you expect black to be for a given sheen?  (Bearing in mind the original file had values around 2 in the deep velvet).

It would make life easier if all questions of artistic judgment could be subject to a numerical answer, but they aren’t. Also, the recommendations are usually negative rather than positive in nature, they say what shouldn’t be done, rather than what should. They say that hair can’t be green, and if dark enough, it can’t be blond either. But they don’t say what it is.

With respect to the darkness and reflectivity of the clothing, first of all we know little about the fabric and so would have to guess. In this context, though, most things that bring out detail in the dress of the woman at right also lighten the dresses of the two at left, possibly to the point of making them seem dark gray rather than black. People are going to have different opinions about how much of this tradeoff to make.

Dan


sj_90000@...
 

Hi Doug,
 
I think your referring to the peineta, which is probably made of tortoiseshell so the brownish yellow color is appropriate. See this Wikipedia entry:
 
 
HTH – Steve
 
Date: Saturday, February 13, 2021 01:30 PM
Subject: Re: [colortheory] Mantillas: Dan's comments
 
Dan, Thanks for such a lengthy review and providing so much info. I will read again and try some of your ideas to see the combinations via blending 2 images.

I do have 1 question. You did not mention it at all that I read; so I'm assuming is didn't seem important to you...but it sure bugged me...Mine was 206.

Mantillas are veils with semi transparent areas we see thru to the background. Yet only a few of us corrected the veils to eliminate the unrealistic (to me) colors in the top parts of mantillas in majority of the images.
So the question: is the better treatment to fix the transparency colors to match the background or accept the bright mis-matched colors....which I agree if they were colorful decorations on the veils would look nice.  But I think the colors are in error???
Can you please comment so we can better appreciate what to do in future veil images; often associated with brides which would seem to need corrections to avoid bright saturated colors in the veils not seen thru from the background.

Doug Schafer


Bill Theis
 

Thanks Dan.  This is by far the most helpful guidance I have gotten in a decade.  Faces have always been a problem for me and I didn't know exactly why.  I went back to my original file and used this great action called "Skin Desaturation" I found in someone's panel & applied before doing LAB color moves and... voila ... things improved greatly.  A small color correction curve afterwards makes to me a satisfactory result.  Knowing the answer (comparison to a much better correction) is invaluable.  And now I know what to watch out for.

There still is the matter of the greenery in the lobby being too blue in the par version IMHO.


Dan Margulis
 



On Feb 12, 2021, at 11:41 AM, Paco <paco@...> wrote:

On 234, went through all the Tests and I can see how much, for the better, these merges make. Now that we know the how. Please explain the why.

That is a very succinct question that deserves an equally succinct answer: you screwed up.

We all screw up from time to time. Occasionally we should know better, as when don’t notice that one of these women’s hair measures green. More often, our eyes adjust to what we’re doing and we decide we have something better than it really is.

How do we find this out? Easy. Enter a case study like this one and check out our work against that of 40 others.

If that doesn’t happen to be a possibility, the best alternative is to check on ourselves. For example, we should test Auto Tone at what we think is the end of our work. For example, I prefer working with a flat capture of the raw file. But I set up the acquisition to also save a version with automated correction. I don’t expect its color to be better than what I come up with, but you never know.

If the image is of any value, and I suspect that my correction has an issue, I’ll make a second one that doesn’t have that particular issue (but maybe a different one) and, if appropriate, blend the two.

If (as was your case) I don’t suspect any problem, then I’ll make a one-minute version. It can be done. I just did one, and timed it and posted it. Here were my steps, and yes, I claim six seconds average for each of them, because in one-minute mode, there is no time for thinking or measuring, you just do the obvious and hope for the best. And a couple of these steps could have been left out without much harm.

1) Auto Color. No time to measure and write curves.
2) Multiply layer because the skin is obviously too light.
3) Add layer mask, RGB background layer, the standard choice, no time to consider something else
4) Blur 30 pixels, ditto.
5) Velvet Hammer script.
6) On to LAB.
7) Looks a little blue still, so shove B curve toward yellow, no effort to get it right.
8) Lasso face and a bit of garment of center woman, and MMM+CB Make instantaneous judgments which appear to have been rather poor.
9) Sharpen script. Remove Hiraloam Darken layer because it normally is bad for faces, not because I’ve examined it.
10) Back to RGB and save.

If you go into the Photos folder for this, you’ll find the result there as #242. It’s certainly not competitive with any of the good versions here, the faces are way too red, the contrast is mediocre, and the clothing isn’t the right color either. But, once downloaded, it can be a canary in the coal mine, letting us know when something is going dreadfully wrong.

If you download it, you can run the tests I did, with the same five versions I used in responding to Bill. Just layer this crappy one-minute special on top of each, and ask yourself, is this as useless as I hope?

On #238 (mine) and #241 (the par) I see this new version as confirming they are pretty good.

Hector’s #232 has yellowish skintone and I suppose that an infinitesimal move toward red with this version would technically be an improvement but it wouldn’t even be worth the mouse clicks.

This version has more color variation in the faces than it should. Gerald’s #223 has less than it should. Layering this version on top might suggest this to him. I don’t know that it’s appropriate to try to blend with this one but it might provoke him to try to make something better.

Putting this version on top of Bill’s #218 immediately reveals color issues; what he does with that knowledge is up to him.

Putting it on top of your #234 shows a serious problem, and it doesn’t take long to figure out that it has nothing to do with the one-minute version’s color. Rather, this is a low-quality repetition of the test suggested in my other post. The luminosity in the one-minute version isn’t as good as Bill’s, but it’s plenty good enough to reveal the bitter truth about yours.

Since we are assuming that Bill’s version isn’t available to you, depending on time constraints you could either blend the one-minute version in at 50% Luminosity, or you could aim for bigger game by producing a blend version yourself.

According to me, the potential benefits of having this canary available are well worth the minute.

Dan







Doug Schafer
 

Thanks for clarifying...makes complete sense now. I probably made an error in my correction.
Doug Schafer


Paco
 
Edited

Yeah well, thanks for the succinct answer Dan. The why I meant is the why do the corrections you made work the way they do. What's happening behind the scenes? Yes the corrections in the test do make the image better but not much more.

I never said that I felt the skin tones in 234 were not good. On the contrary, I see them as being very good compared to the rest of the submissions. All the women have different skin tones but that is normal because of the mix of races. As to the green hair, I measured the women's hair in 234 and as an average I got R 115 G 96 B 96, R 112 G 103 B 105 and R85 G 82 B 87 where G is always equal to or less than R & B. So where do you get more G than I? The skin tones measured as CMYK are almost identical in M and Y except for the woman to the right who is of a more moorish R than the others and then only by 9% more M than Y.

The mantillas have some browns, golds and blues that I missed altogether. Now that I've gone back to the image I wish I had less noise/grain in the dresses and would like to learn how to get rid of it like some others did.

I completely disagree that I screwed up completely and still believe 234 is a very good solution to the problem.

Paco


Dan Margulis
 



On Feb 14, 2021, at 5:56 AM, Paco <paco@...> wrote:

Yeah well, thanks for the succinct answer Dan. The why I meant is the why do the corrections you made work the way they do. What's happening behind the scenes? Yes the corrections in the test do make the image better but not much more.

I never said that I felt the skin tones in 234 were not good. On the contrary, I see them as being very good compared to the rest of the submissions.

I believe that’s approximately what I said too: 

Wow! It’s very unusual to find an image with better color than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. Paco’s hard work paid off. The par is now improved by the addition of his color.

All the women have different skin tones but that is normal because of the mix of races. As to the green hair, I measured the women's hair in 234 and as an average I got R 115 G 96 B 96, R 112 G 103 B 105 and R85 G 82 B 87 where G is always equal to or less than R & B. So where do you get more G than I? 

I have never discussed RGB values. Bill spoke in terms of LAB values and that’s how I responded. I stated, and proved, that while the hue of all five versions being discussed can be described as red, your version falls more on the magenta side than the others, as Hector’s falls on the yellow side, everybody else’s between them, and all five are acceptable for hue.

The mantillas have some browns, golds and blues that I missed altogether. Now that I've gone back to the image I wish I had less noise/grain in the dresses and would like to learn how to get rid of it like some others did.

They didn’t, because they never introduced it in the first place. According to the notes that people submitted, most of the good versions involved some manual procedure that reduced graininess in the faces. I do not recall reading that anybody did anything in particular to kill grain in the garments. Yours does have more grain there than others do and I agree that it should be reduced, because you have smoothed out the faces so nicely that the graininess of the clothing clashes with them.

In that one-minute version I did yesterday I obviously had no time to consider noise reduction, and the faces in it are objectionable for grain, but the clothing is not. So, I would suggest going back in your procedure and finding out where in the process the noise crept in.

I completely disagree that I screwed up completely

My remark that

Paco didn’t say anything that indicated he knew how truly outstanding his color was.

IMHO cannot be read to mean “Paco screwed up completely”.

and still believe 234 is a very good solution to the problem.

I believe that, as submitted, it is better than the majority of entrants. I stated, and showed, how to get it fairly easily to a point where I thought it would have been perhaps the best of all, the reason being,

 ...And if he underestimated how good his color was, it’s safe to guess that he also underestimated how bad his luminosity was. 

Dan


Dan Margulis
 

On Feb 12, 2021, at 10:17 AM, David Remington <david_remington@...> wrote:

Thanks again for the generous and detailed analysis. Mine is the 214 the visitation from the North. I agree with your feedback that the faces are too light and have insufficient modeling and the clothing is a bit too dark. You are also correct that I find some of the other version overdone and not accurate or realistic to my eye. More illustration than photograph. There is a medium in between though. Version 220 comes closer to my taste than 213. This type of work that is really retouching and colorizing is very different from my everyday work.

Everyday, as in, at the present moment, may not be tomorrow, but it’s good that you make the distinction between retouching and colorizing, or, as the figure skating judges would have it, technical difficulty and artistic merit.

So far, we’ve progressed from Hotel Lobby (modest technical difficulty, limited opportunities for artistic originality) to Mantillas (technically difficult, with more opportunities for originality) to Concert on the Beach (not technically difficult, but immense room to make artistic judgments). So, naturally, the fourth one, to be announced tomorrow, will combine that amount of artistic freedom with considerably more technical issues)

Dan


Adrian Thompson
 

Thanks for your very detailed feedback. I might just about scrape enough time to put in an entry for this week!

 

Adrian

 

 

 

From: Dan Margulis via groups.io
Sent: 12 February 2021 03:52
To: colortheory@groups.io
Subject: Re: [colortheory] Mantillas: Dan's comments

 

 



On Feb 11, 2021, at 2:19 PM, Dan Margulis via groups.io <dmargulis@...> wrote:

 

Because my comments on the individual versions are lengthy, I’m putting them in a separate post. This one is the introduction only.

 

And here are the specific comments on the Mantillas exercise.

 

Dan

*****************************************

 

201 This person created much of his facial detail through selective burning, concluding that there was too much natural variation in appearance of the three women for more than rudimentary curves correction to be helpful. The blueness of the garments seems to be a deliberate choice, partly because of concern that the mantilla at right needed a lot of attention. However, as noted in a previous post, this damaged the skintone, giving us the Norwegian appearance.

 

To reiterate past advice: when you think you have finished your correction, run Auto Tone. It should do nothing. If it seems to improve contrast (as it does here) even if we don’t like what it’s doing it’s a strong indication that our work is not over.

 

202 While #201 had a blue cast, this one’s is cyan. These things are easily measurable; see #205 for explanation.

 

203 Fickle fate has dictated that three of our first five entrants have chilly shadows. This one, IMHO, is quite a bit better not just than the last two but than the other half dozen we’ll see later, possibly excepting #205.  The reason? The fleshtones are more correct. Maybe this was due to some of the work this person did with H/S. It probably helped out that he has the dresses relatively dark and the skin relatively light; less chance of them affecting each other.

 

Interestingly, he considered the possibility that the blue dresses were objectionable, and did a second version in which the H-K script was used to neutralize them. He submitted both and told me to choose. I chose this one. The clothing is very dark without losing detail. If the skintone is of a natural hue, then bluish dresses emphasize its yellowness, a plus in an image as generally colorless as this one. As opposed to having an overall blue cast that kills the yellows.

 

If this guy had done some channel blending into the featureless red channel to get a bit more shape into the faces, this one would be right up there with our best.  See #205 and #220 for further comments about this version. 

 

204 Several versions have nice luminosity but questionable color. Here is one that does the opposite, the result of careful curving that assumed that the woman at right was wearing black but that the dresses of the other two might be reflecting some colored light. The faces seem blurrier than our best versions because the eyes and eyebrows have closed up. This may have been the result of some High Pass filtering. So I rate it GOOD COLOR, NO SHAPE.

 

205 Overall the impression is the same as that of #203, with bluish blacks and light fleshtones that are probably accurate for hue. #203 is slightly better, however, due to better shaping of the faces and garments, and the presence of more “irrelevant” color that might convince us that this is not a grayscale image. See #208 and #210 for other suggestions on how to improve this.

 

Four of our first five entries share characteristics, so let’s look at some numbers. I’ve measured LAB values for the central woman in the middle of her hair, and in the dress to the right of the cross. Results:

Central lady’s hair:

#201: 3a(10)b, purplish blue

#202: (4)a2b, green

#203: 1a6b, warm yellow

#205: 2a7b, warm yellow

Par: 5a6b, brown

 

Her dress:

#201: 3a(24)b, very blue, slightly purple

#202: (9)a(12)b, very cyan

#203: (2)a(16)b, very blue, slightly cyan

#205: (3)a(6)b cyan, less so than #202

Par: (1)a(3)b, a cool black

 

Those interested can shuffle color and luminosity layers around among these four. Not having done so, the by-the-numbers prediction is that the color of #203 and #205 would be preferred. Their numbers are questionable, which is better than the other two, which are outright wrong. Further comments about this image are in #208 and #210.

 

206 At first glance this one seems comparable to #203, with dark clothing and light faces. Both have good fleshtones, but this person tried to keep the dresses neutral rather than the blue of #203. Even if you think that was the right call,  I would still prefer #203 because of its better detailing in the dresses. Here the blacks have closed up so much that it’s hard to tell where the women’s hair ends and the mantilla begins. The faces came out well, however. He gave many of the individual components special attention. See #210 and #229 for suggestions on how to improve it.

 

207 Jeepers Creepers, where'd ya get those peepers?

Jeepers Creepers, where'd ya get those eyes?

Gosh all git up, how'd they get so lit up?

Gosh all git up, how'd they get that size?

 

I think from reading the description that they were modified by the program Portrait Pro.

 

208 This person started off on the right track with a luminosity blend of the red channel into the RGB. Thereafter he put in too much color. That might not have been so off-putting if these faces weren’t so full of noise and harsh transitions to begin with. To me it looks like these ladies have beards. A bit of softening as done by others might have made this more acceptable.

 

Nothing wrong with excessive color, provided you use it for blends. For example, blending 25% of this into #205 improves both versions.

 

209 From the late 1990s through 2012, I taught, 32 times, 3-day advanced courses for those who had found that the three-day basic ACT course wasn’t sufficient torture. Throughout that time, this image was part of the curriculum. The classes consistently voted it the most instructive of the 25 or so they had to deal with. And, like you, they found it quite difficult.

 

Our set of 41 entries include four from such classes, between ten and nineteen years old. Hopefully technique will have advanced some in that time. In evaluating them, please be advised that twenty years ago, having a dozen or so layered images of the current size open at the same time put a serious strain on the systems then available. So we worked at lower resolution, and in CMYK. I’ve converted them and upsampled them here. I’ve tried to match the cropping/positioning but it isn’t perfect, so they’re not the best choice for blending.

 

The first such entry is from a 2010 class in Canada. Its daddy was a police officer who specialized in forensic imaging. As that might suggest, he was pretty good at dredging out detail. I am sure he did some channel blending to get that pop in the faces. Colorwise, however, he ran into the same problem many of you did. The woman at left has blue hair, the image has enough of a cool cast to kill the yellowness in the faces, which are too pink. 

 

210 From a class in Italy in 2011 comes this version from our colleague Davide Barranca, who knows what Mediterranean skintone looks like. Since he also is one of those favoring very dark dresses, this effort is most directly comparable to #205 and #206. And since most of us appear to feel that Davide’s treatment of the faces is somewhat too dark, I’d say that blending 50% of it into either would improve them.

 

211 This attractive version addresses the question of what to do with the background. This person’s notes indicate that he realized early on that the lighting was cool, probably an overcast day. He decided to correct it to a sunnier look. He also took into account

 

The women are Spanish, and would normally have some facial color…I assume the wall behind the women is of a church or similar building. Hence the color could be a variety of gray towards yellowish light brown.

 

He did the luminosity blend of the green channel into the red, but it still lacks a little of the depth of the best versions. As for the color, I agree with him about what the background color is in real life, but I actually think we’re better off with something further away from the skin color, possibly a blue-gray. That, of course is merely a matter of taste. Further comments about this one are found in #215, #216, and #221. GOOD COLOR, NO SHAPE.

 

212 This person may have had different objectives than the rest of us. Roughly speaking he was after a period mood, as if was working in Photoshop 3, making it look as the film capture that it is. There’s nothing unreasonable here but I repeat the suggestion that these faces are too pale for typical Spaniards. It isn’t helped by a mild cool cast, measurable in the hair of the two women at right.

 

213 When I first saw this version I under-rated it. On further review it seems to me one of the best entrants. The procedure was straightforward: a luminosity blend of the green into the red, then curves and Channel Mixer to neutralize the clothing, Davide Barranca’s ALCE sharpener, and if I read the documentation correctly, a curves correction with the faces selected. That may have been what threw me at first, it seems to me that the dark areas (the eyes) are lighter than corresponding points elsewhere. And all three of the chins have gotten quite rough—softening needed. There may have been too much MMM in the faces, but on the whole, a good entrant.

 

214 Superficially this one is similar to #206, very dark clothing with very light faces. But at least #206 threw some color into the faces. This one is a portrait of three ice-women. The treatment suggests that the person wouldn’t like something as aggressive as #213. Still, blending 50% of it into this one would give it more life.

 

215 Out of the refrigerator and into the deep freeze. This person made some sophisticated blending moves and then, as I read it, had the same blueness in the clothing that plagued several people. He was wondering about how to get rid of it when the devil, who never sleeps, whispered into his ear that the solution was to blend the red channel into the blue, but fiendishly refused to specify Darken rather than Normal mode. That neutralized the clothing, all right, but it took the fleshtones along for the ride.

 

The color is enough to tempt one to hit the delete key. It would be a big mistake. This is just another example where the luminosity is better than the color. This is merely the most extreme example—so far, just wait till the next one. The color, instead of just being so-so, is terrible. OTOH the luminosity is among the best in the group. So, if you take the luminosity of this one and combine it with the nice color of, say, #211, the result is better than either. GOOD SHAPE, NO COLOR

 

216 The horror movie continues but with a similar happy ending. For the color it’s enough to quote the same numbers as in the comments to #205: for the center woman’s hair, 7a(13)b, strongly purple; and for her dress (5)a(30)b, a brilliant blue tending toward green. In short, casts far more extreme than in the group discussed there.

 

How to fix it? Start over. But don’t throw this version out, no way, because it has outstanding detail, maybe the best of all 40 entries. He used the green channel blending into the RGB, but especially he did some fine handwork with dodging and burning to get outstanding shape in the faces.

 

So, set this version aside, and make a new one where good color is the only issue. It’s quite liberating not to have to worry about channel blending or sharpening, etc., and one should be able to whip a good one up in a few minutes. Discussion of this version will continue at #234. GOOD SHAPE, NO COLOR

 

217 Another from my advanced classes, this one from 2002, executed by me. For twenty years old, it holds up rather well. Alas that it doesn’t have enough resolution to be a true entrant in this study.

 

218 For the third time in the last four images, the person has lost control of the color but left us with excellent luminosity. This image has a magenta cast as measured in several places. The most objectionable feature, however, is the variation in the faces, going from speculars, to yellowness, to mottled reds.

 

The excellent detailing is due, first, to the green blending, but also to a side trip into CMYK, where it’s easy to isolate and work on black objects like these dresses. As a result, he has maybe the best shape in these garments of anybody. There will be further discussion of this version at #234. GOOD SHAPE, NO COLOR.

 

219 A purple cast wipes out the yellowish nature of the fleshtone. The hair of the woman at right, for example, measures 7a(14)b, heavily purple. This should be checked routinely as part of the correction process.

 

220 Chosen for the par version. The person didn’t enclose a list of steps, so I can’t discuss technique. What I can say is that he is one of those who believes in very dark blacks, but unlike previous such versions, his faces match them appropriately. Don’t believe it? Compare it to #203 and #206, where the garments are similar but the faces much lighter.

 

221The color is basically correct, but the person used channel blending to neutralize all blacks. It therefore compares unfavorably to #211, with which it is superficially similar. For example, the woman at right’s hair is an attractive brown in that version and black here. Also, more agreeable color transition in the dresses.

 

222 Fear of a cast, as it did in #215, leaves a picture that’s too neutral. This person was aware that the ladies are Spanish and treated their skintone accordingly. But his curves must have knocked out some of the saturation in the faces as well as the blacks and the background.

 

His sharpening was sensitive, excluding the skin and also the background people at right. But he then desaturated the people at right. That’s the opposite of what many of us, including me, did. When I see a picture like this, it almost seems like a grayscale. Since not much can be done elsewhere, if I had to improve this version in a few seconds I’d paint saturation into the lips, the red necklace, and these people at right. This person was worried that they’d be too distracting, but I think they’re too small for that.

 

To prove that the basic color of this version is correct, even if too gray, put the par version on a layer above this one and set it to Saturation mode. Then, change the mode to Normal, which should reveal that this current version has somewhere along the way lost desirable detail in the eyes and eyebrows.

 

223 Chosen for the par version. The detailing is good, and the color is satisfactory, because by the numbers everything works. Gerald Bakker, however, hit the nail on the head

 

I assume that facial detail is the most important aspect of this exercise. I myself focused mainly on that. [Mine] looks remarkably similar to the par version, except for one thing: the redness of the lips. I emphasized eyes, teeth, jewelry, but forgot the mouths. 

 

We can go further. The original is so lacking in color that it could almost be taken for a grayscale. To prevent boredom, the experienced person finds an excuse to shoehorn some color in, the lips being an example. But it’s hard to put much extra color in the faces, the hair, the background, or the clothing. So in the par version, there is also additional color in the little boy at right and in the women’s necklaces. Coupled with slightly shapelier faces, and I would say: if you happen to like Gerald’s version, then it would be impossible not to like the par version more.

 

This falls in the category of those with excellent luminosity and color that—while not incorrect—can best be described as “good enough”. I’m sympathetic because I had the same problem. After two failed from-scratch efforts, I decided that I had too many balls in the air at one time, and to create one version for luminosity only and a second for color only. And I then created a third for color variation (as in the lips), used for masking in to the final product. GOOD SHAPE, NO COLOR

 

224 No steps were enclosed here but it looks like this person fell into the same trap as #222: the garments are extremely neutral, and he probably tried so hard to achieve it that he also neutralized the faces. Since he likes very dark dresses, it would be well to compare this one to #203 and #206 to see how important it is to get some kind of color in the women.

 

225 This person acknowledges having a difficult time here. The dresses get a nice middle-of-the-road treatment, but the faces are too light and too colorless.

 

226 Too flat, and shadows rather plugged, despite having colors that are basically correct. I once again suggest to all that when you think you’re finished, you run Image: Adjustments>Auto Tone. If you’ve done good work, it shouldn’t change things at all, let alone the major change it makes here.

 

227 In 2010, eight years after having done #217, I tried again in another class. In the interim, MMM had been introduced, which may account for the better color variation in the fleshtone.

 

228 This one is quite close to #223, for which I have a mild preference. There’s good shaping in the faces here but they are rather noisy. I also personally think that the darker clothing in this version makes the faces look weaker, even though they’re so similar to those of #223.

 

229 I believe the problem is that this person, instead of ending with an experimental Auto Tone, started out with it. As a result (I surmise) there are large nearly blank areas in the faces, marring an otherwise acceptable treatment.

 

230 We’ve seen quite a few of these before. Distinct blue cast. Using the same sample points discussed in #205, the center woman’s hair measures 3a(5)b, purple, and her dress is (3)a(17)b, very blue tending to cyan.

 

As with the others, the blue cast damages the inherent yellowness of the skintone and makes the women too pink.

 

231 Chosen for the par version. The skintone should have been softened somehow. It lacks the smoothness of, say, #223, but the variation makes this the more interesting treatment. This person was interested in getting more detail into the dresses, particularly the one at right. So he did some neat CMYK trickery with two different GCR settings. CMYK is the best way to handle detailing in black objects because they reside in the black channel where there usually isn’t much else to fork up. The usual argument against going to CMYK when the final is to be RGB is that it can eliminate certain colors that are within the RGB gamut but not the CMYK. Obviously, there aren’t any such colors in this original.

 

232 Chosen for the par version. By Hector Davila, this is my favorite, with a very appropriate skintone, no obvious problem, and a generally conservative treatment. A couple of interesting notes: he was (I think) the only one who wheeled out the ancient Color Balance command. Maybe it’s appropriate here, because its structure (Highlights, Midtones, Shadows) lets it more or less isolate the background, faces, and garments. Also, many of us noticed that the woman at right seems darker than the other two. Hector was (I think) the only one who tried to compensate by selecting the other two women and correcting them individually.

 

233 This is very comparable to #222 except that here the clothing (and hair) is darker. That would be a matter of taste. Otherwise, the two share the same problems: overly neutral skin and unsatisfactory treatment of the eyes. 

 

234 Paco Márquez gave a complete description of what he did here in a post on 10 February, so I won’t reiterate most of his steps, but his basic philosophy was:

 

Skin tones I worked on separately because here you have the woman on the right as a rosy one, the next one with a more Moorish tone and the other 2 as a mix of both. In Spain, because of the mix of races, skin tones are very varied. 

 

Accordingly he spent much more time than most working on the faces in isolation. Yet at the end he wasn’t wholly satisfied with the result.

 

This is a good thing. This is why we all go through these exercises. You submit an image you think is good, and it turns out not to be so. You don’t need me to say it outright because it’s obvious by comparing yours to a better version.

 

It’s not enough, though, to just say that the other is better because it’s easy to pick the wrong reason. In this study, many people think they have inferior versions because their skintone is too pink. Sometimes that is the correct answer in the sense that the rest of the image is OK. More frequently, they’re mistaking the symptom for the disease. The problem is that the version has a blue cast, which kills the yellow component of the flesh. 

 

Similarly, Paco’s self-evaluation:

 

I missed the subdued colors in the mantillas which show up well in the par version, and screwed up by desaturating them too much. Apart from that I feel I pulled out a credible correction of the original. 

 

I think that this again mistakes symptom for disease, and hereby offer a demonstration that I recommend to everyone, because it says a lot about how to handle this image. Here we go:

 

Gather the ingredients:

2 copies of the par, #241.

1 copy of Paco’s #234.

1 copy of the horrendously blue #216.

1 copy of #218, with its swarthy skintone.

 

FIRST TEST.

1) Add a duplicate layer to #234.

2) Set it to Color mode. Appearance doesn’t change because the two layers are identical.

3) Shift-drag this Color layer onto one copy of the par.

 

Wow! It’s very unusual to find an image with better color than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. Paco’s hard work paid off. The par is now improved by the addition of his color.

 

The catch: Paco didn’t say anything that indicated he knew how truly outstanding his color was. And if he underestimated how good is color was, it’s safe to guess that he also underestimated how bad his luminosity was. Therefore,

 

4) Change the Color layer’s mode to Normal, and toggle back and forth between the two modes.

 

Disaster! The drawback Paco pointed out was only one of many. Best approach at this point: start over, and make a new version, worrying only about contrast. It may make it easier to actually produce a grayscale. And why not? The color problem is already solved.

 

SECOND TEST.

1) Add a duplicate layer to #216.

2) Set it to Luminosity mode. Appearance doesn’t change because the two layers are identical.

3) Shift-drag this Luminosity layer onto the other copy of the par.

 

Wow! It’s very unusual to find an image with better luminosity than the par, let alone when that image isn’t good enough to be selected for par itself. But that’s the case here. 

 

4) Change the Luminosity layer’s mode to Normal, and toggle back and forth between the two modes.

 

Disaster!

 

We know that two wrongs do not make a right, so there is no point of blending the color of #16 with Paco's luminosity, unless we are trying to make mockups for a horror movie. However, the question is whether two rights might make a wrong. The logical prediction would be that the luminosity of #216 and the color of #234 would make a dynamite pairing, but...

 

THIRD TEST.

1) Shift-drag the Luminosity layer from the second test onto #234.

 

Not bad, but perhaps not as good as expected. The shape of the faces is great, but the color isn't what Paco had in mind IMHO. His color depends on faces darker than these. I'm not saying that the faces of #218 are too light in isolation, just that they're too light in combination with Paco's color choice. Similarly, the clothing of #218 is much lighter than in #234. So,

 

2) Reduce the opacity of the Luminosity layer to 50%.

 

I see this as an improvement over #234 as submitted. But, we must concede that it saves 50% of Paco's original luminosity, which is 50% more than deserves saving. Therefore,

 

FOURTH TEST.

1) Restore the Luminosity layer to 100% opacity.

2) To it, apply #218, (replacing #216).

 

I don't think #218 is quite at the level of #216 for quality of luminosity, but it's pretty good, and is much more in line with Paco's views on how dark to make the faces and clothing. Therefore, 100% opacity is just fine.

 

3) Get rid of the noise in the faces that #218 brought along for the ride.

 

And there you have it. I find this combined version to be preferable to the par, even though neither #218 nor #234 as originally submitted is worthy of much notice. This is why I wound up doing my own version this way, except that it did not come out as well. Let us remember, therefore, that there is no disgrace in evaluating a certain version as GOOD COLOR, NO SHAPE.

 

235 Similar to #230, except bluer and flatter. Once again: run Auto Tone at the end. The failure to do so accounts for much of the flatness here.

 

236 This person tried a lot of things before becoming frustrated. This resulted in a hybrid of issues. The dresses morph into a deep blue that is disagreeable. Yet it isn’t a true blue cast, because the skintone isn’t affected much. It’s too light, and needs some noise removed, but that’s another story.  Here I’ll just describe the blend that fixes odd blue areas.

 

On a duplicate layer, apply the blue channel to the RGB in Lighten mode. This works because the blue is commonly darker than both others, except in blue objects, and notably in lipstick, where the green is usually the darkest. And a few other exceptions that are not found here.

 

This move completely neutralizes all blues. I would think that’s inappropriate for the boy at right, so I’d add a layer mask and paint his color back in. Also, the blend completely neutralizes the blue areas in the dress, a small amount of which might be desirable. So reduce layer opacity to taste. And then the picture is more respectable.

 

237 The image has a mild blue cast, not as bad as some of the others, but still presumably is partly responsible for the Scandinavian-style skin. Here the skin has been softened to eliminate the noise, but shape was lost in the process. To get an idea of how it might be improved, try blending in #216, in Darken mode to preserve the blacks of this version. Setting the blend layer at 50% Luminosity should get a decent result.

 

238 Chosen for the par version. It’s mine, and if you read the comments on #234 you know what I did.

 

239 This one is quite similar to the par version—except for the overly light, overly pink faces. We’ve seen several version where the fatal pinkness was caused by an overall blue cast, but that isn’t the case here. I’m not sure how it came about.  Everywhere I measure in the skin the A value comes out significantly more positive than the B. With Mediterranean types the reverse is usually true, except where they’re wearing makeup.

 

Whatever put the pinkness into the faces also put it into the whites of the eyes. I measure those of the center woman as nearly 20a. It’s true that these areas should be somewhat pink because of the blood vessels in the eye, but they shouldn’t be as bad as this. Making them more neutral makes the surrounding skin more palatable.

 

240 The original gave us a lot to work with in the clothing but pitifully little in the faces. Accordingly, our failures were almost always due to inadequate treatment of them, or issues of color that did bad things to them. As best I recall this is the only version with a big problem in the detailing of the clothing. Everything is closed up. To get an idea of what’s missing, grab a copy of #231, where the guy went to a lot of trouble to get detailing in the blacks. Blend it into this version, 50%, Lighten mode. That won’t affect the faces, but it will show how much better this version would look with more definition in the blacks.

 

241 The par, blending the versions designated above, with each given 20% weight.