High Andes: Dan's comments
Hi everybody, I'm Fabrizio Cacopardo and I'm italian.
I entered this group not long ago and this is my fourth entry, n° 1022.
First thing first, I'd like to thank Dan for being so generous and always dedicating his time to our work, commenting every single image. A demonstration from him is worth more than a hundred tutorials on Youtube.
I learned so many new techniques within the last month and I truly believe the comparison with others to be essential since it helps me realize what my limits are, hence I thank you all too for your precious feedback.
Wishing everyone a nice day,
In esthetic choices, there's often both reversion to the mean, and recourse to authority (group think). Your cover story shows your anticipation of these effects. But there's also oppositional posturing, which happens most often when consensus produces something especially bland or ill fitting the use. Again, I'm mainly in fashion, and there's both meaning and a random walk in trends, and then the random is granted meaning, which builds. Apologies that I'm circling around a question that remains ill-formed. As an example, back in the day of "Professional Photoshop" and "Makeready," I hit my endpoints and obsessed about the dropper readings. The fact that photoshop 3 didn't have multiple dropper points like a scanner interface was so annoying that I wrote cranky letters to Adobe. For the purposes of these studies, I generally got close to full range, but in paid work, I aggressively float the ends and try to fix it with contrast. I still flip through channels after I spot the file to see if there's a problem, but I rarely use the dropper unless I'm matching seamless. I'm trying to leave the mistakes in. If I see a cast, I'll push it a bit farther. I do what I can to make the picture just subtly wrong enough to slow down its ingestion. It seems right now, but that would have felt crazy back when everyday was a struggle to keep an LVT, E6. drum scanner, imagesetter, rapid access line, and contact proofs linearized.
Regarding Franklin, I recall reading a story of a client of his complaining to him about Caslon and how it was inferior to Caxton, to which Franklin agreed, and showed him a specimen of something in Caxton he was currently printing, which the client liked, which was of course Caslon.
I suppose it depends on the crowd. If referring to group members, we’re hardly typical of the general public. We have biases that they don’t, typically because we know what tricks we are using and because we can see their telltale signs, we assume that laypeople can, too, and that it will bother them. So we’re more inclined to reject for what we consider oversharpening, or for colors being too loud, than they are. OTOH they are quick to punish those who don’t have a full tonal range.
Preferences between versions is partly a matter of taste but a lot is technical and artistic proficiency. When I was teaching ACT up to eight people would submit versions and then we would decide which were the best one(s). Approximately half the time the vote would be unanimous for a certain version as best. Not surprising, since half the versions would have disqualified themselves.
And that’s the case here. You and Gerald often post lists of your favorites before knowing mine. Granted, all three of us have very different tastes generally, yet usually any two of us would agree on three of the top five. That wouldn’t happen if we had 30 or 40 excellent entries to choose from. Typically, however, only eight or ten could realistically be considered, so even if we each selected our choices by flipping coins we’d still have a lot of overlap on our lists.
The question is also how a group would determine its preference. Is it the version that gets the most votes? And how do we adjust for images that some like and others hate? That’s the virtue of the par, we can criticize it but nobody is really going to dislike it.
More commonly a group tries to reach a conclusion by consensus, which usually results in a bland choice. As Benjamin Franklin once said about our discipline, “If all Printers were determin’d not to print any thing till they were sure it would offend no body, there would be very little printed.”
Something along those lines is described in the final pages of CC2E. Book publishers traditionally consider the cover artwork of supreme importance, and probably it was in a time when people actually went to bookstores. I imagine that 95% of those buying CC2E did so on the basis of a thumbnail online, but that did not prevent the entire staff of Peachpit Press from getting involved in the decision and spending hours and hours on it.
First, they investigated stock photography of canyons, without checking with me first. I suggested that I know some pretty good photographers and that I have quite a selection of canyon photography to choose from, and I myself have a bit of experience in preparing images for CMYK. They asked for samples, and I gave them a dozen, all tagged sRGB. By whatever method they were using they narrowed it down to two, and they sent me high-quality proofs of their designs for each one. The images were quite a bit louder than I had expected, since the art director knew nothing about color management and had set up her system to open all files in Adobe RGB, ignoring any embedded profiles. After discussion, we selected one, but they requested that I submit a few color-corrected-for-CMYK versions of it so that they could make a more informed decision.
At about that time the beta readers were getting rather uppity about the quality of some of my corrections elsewhere in the book, and I suggested to them that if they were such hotshots maybe they’d like to prepare their own versions of the cover art, now that a certain original had finally been chosen. And I said I would send the best results off to Peachpit, without identifying them or expressing my own opinion.
What followed was something quite similar to our case studies: nearly a dozen entries, followed by a vigorous discussion that got even more vigorous when I announced I wasn’t going to submit one of the group’s favorites because I didn’t like it, and since it was my book, by God, I’d veto any entrant I pleased. But I okayed the group’s other favorite, an Italian entrant. I also submitted two of my own, and that of one other beta reader, and I said I was told that all four would be proofed, and then six people from Peachpit would make the decision without help from any of us. The group, but especially the Italian delegation, predicted theirs would be chosen. I named the more exuberant of my two versions as the likely winner.
Then, the apocalypse. The art director to whom I submitted the four unilaterally vetoed the Italian version as being “sharpened too much”. (It was no such thing, it merely assigned contrast to an area that bothered her.) She proofed only the other three and sent them out for a final decision, nearly causing an international incident when the beta readers heard what she had done. I then revised my prediction. I said they would now pick the more conservative of my two versions.
And that proved correct. Groups like these often split the difference. If all four had been submitted, from most to least conservative they would have been the fourth version; my two; and the Italian version. I thought that they’d see the Italian version as the most aggressive of the four and would choose my less conservative one as a substitute. But with the Italian version removed from the competition, they would see my predicted winner as the most aggressive, and would pick my conservative version instead.
Bottom line: the question is rather complicated, whether experts or laypeople make the decision.
Once again my compliments to Dan for taking so much time and effort to help us all learn and do better; via many different images. If learning wasn't so much fun, I might have gotten frustrated trying to solve so many problems.
The "real" problem is not achieving the results technically; but rather understanding, analyzing, and deciding what is important, what needs changing, and what colors, contrast, luminance, etc. would be the desired result.
And one more new learning... While usually blended results (to get a par image) invariably yields a better result than any individual image, all blended images are not equal; and some can be less desirable. So it is back to the basics every time: analysis and making the right decisions; then technically executing a good result (and checking the result to verify it matches the desired end state). If enough time available; make an additional correction and blend the two (or more) and assess again.
These have been quite instructive exercises for me. Mine in this exercise was 1003. I do wish I’d toned down the luminosity of the face a bit but otherwise am happy with the outcome. I need to go back now and go through a number of the exercises and rework them in light of the suggestions given to us by Dan. A number of his demonstrations have certainly opened my eyes to many possibilities.toggle quoted messageShow quoted text
Looking forward to the 11th exercise in a week or two. Can’t thank Dan enough for his efforts and the imparted knowledge he’s given us.
, with purpose and great thought.
On Apr 10, 2021, at 3:45 AM, John Gillespie <john@...> wrote:
Another comprehensive, instructive and entertaining lesson from Dan, it will take a while to absorb all the various points and techniques listed.
I have not previously explored the various sub-components of the Sharpen 2018 action, the example in 1015 was enlightening, as was the detail on the Colour Boost 2018 in 1018.
For each image I have a fleshtone value, taken with 101x101 pixel average, taken to our right of the nose and below the eye. I think I was expecting a B value 3-5 higher than the A. But the two pars are only 1 and 2 points higher, and I still find them a little yellow. I note that my own favorites actually have the A higher than the B. Why we should prefer a redder, or a more magenta look may have something to do with the bluish, high-altitude lighting, or it could be because the brilliant red clothing makes the flesh seem yellower and we have to compensate.
I've determined and indicated the top three entrants in four different categories: 1) lightest skin; 2) darkest; 3) most magenta (more accurately, most unbalanced toward A); 4) yellowest (most unbalanced toward B). I don't think it's a particular honor to get one of these awards; they suggest that your views are out of line with most of us.
The demonstrations below produced four new par versions. The new versions, #1028-1031, are in the Photos folder, or you can download all six existing par versions from our Files section, search for
As noted earlier, I wasn’t thrilled with at least one of my choices for the par version, so I chose a different five for an alternate par version, #1027.
1001 (65L 29a 27b) Chosen for the par version. This person did not channel blend; instead he used a very heavy application of MMM Luminosity to bring out the shape of the face, first having gotten accurate color in RGB, although he later discovered that the hair (like others) was in principle too cool, which he corrected.
A couple of quibbles. This person used Bigger Hammer isolated to the bill of the hat, to bring out its detail. I don't know why he wanted it. This is not an important item and we should be careful about drawing attention to it. As long as the area isn't blown out completely it's OK with me.
Also, this person understood that sharpening the face could cause difficulties. Therefore, when he ran the Sharpen 2018 action, he cut opacity to 50%. After finishing his work and reviewing past case studies, he decided he had underestimated the need for sharpening in the clothing, so he added that separately.
Certainly we want to sharpen the eyes, but I see no reason to sharpen the skin at all. In this image there is no difficulty applying a heavy sharpen to the clothing and none at all to the skin: it's a simple Blend If in LAB, because the reds in the clothing are far more A-positive than the skin is.
This version is used for a demonstration at #1021. I had also, in a previous thread, contrasted this version to #1101 for the difference in sharpening. There is another mention of this image at #1009.
1002 (82L 18a 18b, the least colorful of all; also the lightest skin by a wide margin; winner of the Bram Stoker award for the entry that most makes us think that a vampire has been present.) Bill Theis:
In my submission #1002 witness the rare "100% native blood Latin American" Scandinavian. In my zeal to avoid making her swarthy like I did for the Mantillas, I lightened the entire face dramatically. I also accentuated the dirt on her face which I found charming (the exact opposite of #1013 which improves mine when blended at about 30%). I know the exact curves step where I did this as I wrestled with just how light and what skin color was desired neglecting what her true darkness and color might be. In the end I guess I made a Caucasian, one with a filthy face. I think a couple of others fell into this trap (I'm looking at you, #1024) but others went for very, very dark. Lots of blends are possible....
1003 (73L 20a 25b, the third lightest skin) Chosen for the alternate par. An attractive effort of the lighter emphasize-all-reds category. The process notes state:
Yes, many people had the same problem. But most found a partial solution and therefore got a better shape in the face. The key is getting a better red channel, normally by blending.
To get an idea of the importance of that, go to #1027, our first alternate par. The two images seem similar, #1003 is slightly more saturated but the face seems somewhat blown out by comparison. So, substitute the red channel of #1027 for that of #1003 on a duplicate layer. The change is remarkable, though it might be best to change mode to Luminosity. Another comment about this image is at #1009.
1004 (63L 22a 15b, the second most magenta and the second least colorful skin) David Remington:
My version appears cool and dark in the context of most other entries. Looking at it layered with the par I can see blending in a little color, Just a touch though maybe 20%. I can go either way with a blending in a little luminosity. I find the par and many other version too orange in the child's skin tone and in the red stipe of the fabric. I was aiming to avoid making the skin tone too warm or too saturated. In my base export from camera RAW the clothing was very saturated and there was little separation between the orange and red stripes. I increased that separation and reduced the saturation quite a bit. I'll think on this one a bit more and am interesting in reading Dan's take and more thoughts from the group.
My take is the the contrast is fine, slightly darker than the majority, which is a matter of taste. The color is not a matter of taste, it is outright wrong and should have been detected and corrected.
The skin values measured above are rarely found in anyone whose natural hair color is not blond, although some Caucasian children might have it as well. Given the strange lighting conditions, and that certain indigenous people were once called redskins for a reason, I am not prepared to state absolutely that this value could never be right in this scene. But it's a suspicious reading, and the sensible thing is to look at other objects, even insignificant ones. Here's what I'd look at:
*Hair. 5a 1b in bangs, 4a 1b in shadows. As with the fleshtone, I can't say definitively this is wrong what with all the possible reflections, but it's nonstandard and suspicious.
*Stripes. As David notes, there is not good separation between the redder and oranger stripes at top. This is what might be expected if the image is deficient in yellow.
*The turtleneck or whatever it is at right, 12a (6)b, an unusual color for garments in most cultures. Suspicious.
*The approximately neutral background at left. 5a 1b in both the lighter and darker areas. Very suspicious, as magenta is the last hue we'd expect here.
*The brim of the hat. Generally 4a0b. Wildly improbable. It could be neutral, warm, or even bluish, but never magenta.
*The dark eyes. 15a8b. Impossible. The eyes are in context obviously dark brown. The B cannot be less than the A.
The cast is therefore confirmed by multiple observation. It can be reconfirmed by finding any version with more equal A and B--even something as disagreeable-looking as #1002--and blending it in Color mode to #1004. That demonstration should convince us to make a quick, better equivalent of #1002 and figure out exactly how to move away from the cast without damaging the desirable features of the version. Another mention of this version is at #1020.
1005 (69L 41a 46b, the most colorful skin of all by a large margin) This isn't a good time to make the skintone so radioactive, as it detracts from the redness of the garments.
1006 (64L 29a 27b) Chosen for the par version. My personal favorite other than various blends. Ken Harris has posted about his objectives, which were carefully planned, mostly subtle changes after getting the basics right during acquisition. If you like it as much as I do, note that he is somewhat on the magenta side of neutrality, the opposite of #1001 and #1006. Also, his dark hair measures 7a 2b, almost as strongly magenta as #1004 (which is why in those comments I said the value was possibly acceptable) but without the obvious overall magenta cast.
Another comment about this version is at #1018.
1107 (66L 26a 32b, the third yellowest skin) Chosen for the par version. Gerald Bakker planned from the outset to do two versions, to be combined later. He used H-K to emphasize the brighter colors and his MMM action to get excellent definition in the fabric. He writes:
In a previous post I compared this one to #1101, and another comment about it is at #1009. The business about how to handle grime is in a demonstration at #1019.
1008 (66L 26a 32b, the second yellowest skin) Subduing and darkening everything else makes us focus squarely on the face. All good so far, except that the image is now too flat, as can be seen by running Auto Tone.
If, as here, Auto Tone causes a major change, it doesn't mean that you should accept it and hope for the best, but it does mean that the work is not over. Here, getting that extra lightness into the hat and the white woven areas of the garment, would be an improvement even if the face is excluded, and probably other areas would benefit as well.
Other comments about this version are at #1015 and #1022.
1009 (72L 28a 20b, the most magenta of all) Chosen for the par version, which in retrospect was a mistake. Edward Bateman:
I had painted myself into a corner by choosing #1001, #1006, and #1007 for the par. They have too much in common to make a decent par version. Given that my fourth choice was quite dark, I wanted a lighter one to make up for it, and chose this one, disregarding the disagreeable parts of the face. #1103 would have been a better choice. I should also have flipped a coin and discarded either #1001 or #1007.
In #1019 there is a demonstration on how to proceed to tone the grime down, which would be useful here as well.
1010 (62L 28a 22b, the third most magenta and tied for the third darkest skin) Christophe Potworowski made a modest attempt to clean up the face, understanding, he writes, that this is not the cover of Vogue. He adds:
I also had difficulty finding a neutral or white point. I picked the white of the eyes, but this was a mistake, I should have picked a reasonably clean spot on the white hat as per the par image.
Beware of trusting adjectives of color in everyday language. Whether we say red wine, vin rouge, rode wijn, Rotwein, красное виноor, or vino rosso, it's not red but purple. And white wine is normally green, and black people and white people are both red.
Similarly, the whites of the eyes are actually pink, because of the presence of blood vessels.
When you don't have true color look for impossible ones. Hard to say that the face is impossible, but the hair measures blue-purple: 6a(3)b. Brim of the hat is (9)a(6)b. That can't be right. This coolness is what's causing the lack of a break between the orange and red stripes at top of garrment
The blueness causes there to be less of a break between the red and the orange stripes at top of the shawl, among many other problems.
The luminosity of this image, however, is probably better than the par.
1011 (66L 24a 29b) Chosen for the alternate par. John Furnes:
This version is also mentioned at #1020.
1012 (74L 26a 24b, the second lightest skin) Robert Wheeler teaches the value of simplicity. He spent a lot of time trying to get decent color in LAB, gave up and trashed his efforts, and found he could do passably in RGB:
1013 (67L 38a 34b, the second most colorful skin, largely because the neutralizing dirt has been retouched away) Well, this is one way of dealing with the problem of the dirt etc. in the face, but it is not the way the rest of us see it, There is room for debate on how much of the imperfections to show, but none that its presence is critical to understanding the image. This version is used in a demonstration at #1019.
1014 (62L 30a 35b, tied for third darkest) Chosen for the alternate par. John Gillespie's process notes:
And after everyone’s work was posted, he added:
I have more contrast in the garment and less in the face than the par - I suppose this is better than the other way around. I did not want to over-emphasize the dirt on the child's face, which I think the par does a little too much.
It's apparent from the threads that many of us feel the same way.
1015 (66L 23a 19b) Chosen for the alternate par. This is my own version. I started with the flat original and corrected it slightly because it seemed that both highlight and shadow measured too cool. After all sorts of legerdemain, I forgot to re-check those points, and left them cooler than they should be. Nevertheless, blending this one in at 50-50 seems to me to make the par better.
I tried out some uncommon tricks to produce this:
*I immediately selected the eyes, and on a luminosity layer, blended the blue into the RGB in Lighten mode.
*I used a Channel Mixer move that is described in #1016.
*I realized that trying to sharpen the green channel, or its equivalent elsewhere, would be a waste of time, so I decided to take the file into CMYK and sharpen the black, and later sharpen the red. I did use the Sharpen action while in LAB, but only for hiraloam effects.
*While in CMYK, I added weight to the black. This is the equivalent of running H-K.
*While in CMYK, I lightened the midtones of both the magenta and yellow channels. This was in lieu of running MMM later, and may have been more effective. I did, of course, run Color Boost in LAB.
*When running the sharpen action of the PPW panel in LAB, I jacked up the Hiraloam Color control. This resulted in better definition in the narrow stripes of the garment and in the individual threads hanging from her shoulder.
This is one of three versions that seem to be making a conscious effort to get darker reds. The others are #1008 and #1022.
1016 (71L 33a 35b) This one has a good face, but is unacceptable because of lack of detail in the reds. Hector Davila doesn't address that point, but has another good one:
Well, I know how to do it, and how to brighten the yellows and add snap to the reds at the same time.
Ingredients: #1016 only
1) Add two duplicate layers, but make the second temporarily invisible.
2) To the middle layer, apply Channel Mixer to interchange the red and blue channels (Red=100% blue Blue=100% red). This changes the clothing, the face, and any other red object to blue, and changes anything originally blue to brown.
3) Change blending mode to Lighter Color. All the red stuff is now restored, because the red channel carries greater weight than the blue. Making the shawl change from red to blue, as step 2) did, also darkens it, so the blend mode excludes it.
4) Now the red stuff is unchanged; the blue stuff is lighter but has turned brown. So, turn on and change mode of the top (third) layer to Hue, and the blues return, lighter than in the original.
5) All of the above did nothing to increase definition in the reds, to which we now turn. Make a Composite layer based on what we have now, but make it invisible.
6) Duplicate the second (Channel Mixer) layer and move it directly below the composite layer. The layer order from top to bottom is now: Composite (invisible), Second Channel Mixer; Copy of Original in Hue mode; Channel Mixer layer in Lighter Color mode; Background
7) On this second Channel Mixer layer, apply a curve to the L by moring the highlight point so far into the center that the face is wiped out completely. It looks ridiculous as is, but it does get more detail into the fabric.
8) Change the mode of this layer to Luminosity.
9) Turn on the top (composite) layer. Go to Blend If and either exclude everything that is strongly A-positive on This Layer, or else strongly B-positive. Either setting will permit the better reds from the Second Channel Mixer layer to show through. Using the B slider will also lighten the yellows. This is what I chose.
10) Since the bright reds are now a bit rough, reduce the opacity of the Second Channel Mixer layer. I chose 75%.
The result is 1038 and shows, I think, that Hector was right in wanting to lighten the blues. This version might satisfy those looking for a brighter overall appearance than provided by the other pars. Another comment about this version is found at #1022.
1017 (56L 23a 28b, the darkest skin of all) To see what's wrong with this one we need look no farther than the red channel, which is key to contrast in any red objects, including the face. Compare this one's red channel to, say, that of one of the pars.
1018 (59L 25a 20b, the second darkest skin) Chosen for the alternate par, more or less as a counterweight to #1003, which I find too light. I am not a big fan of #1018 as submitted, but apparently I have company. Harvey Nagai:
This makes a good point: if the skin is as dark as presented here, then the color is satisfactory, but if it gets lightened some more color is going to have to be added or the whole thing will look boring.
Ken Harris's suggestion:
On 1018, kicking the endpoint white over ~26 straight and adding 6 points yellow makes an image better than either par, if you ignore the grey and the white, which are areas easy to mask out of the cc. That gives you all the reds, better skin tone, very natural, very NatGeo.
That works, but I'd prefer the flexibility of the Color Boost 2018 action, which is in the bottom half of the PPW panel. It's designed for situations like this with multiple variables. Here there are at least three: how much to lighten the face; how much color to add; and how much, if at all, to move away from magenta and toward yellow. Understanding that the following has many settings that the user can change,
Ingredients: #1018 only.
1) Convert to LAB.
2) Apply the Color Boost 2018 action from the My Actions section of the PPW panel
3) By default this puts us on the Endpoint Adjustment curves layer. Open the L curve and move the highlight point inward from, say, L=100 to L=88. This lightens everything and blows out the hat and part of the nose.
4) In Blend If, choose Lightness, Underlying Layer; move the slider away from the highlight to restore the lightest areas, then split the slider with Option-click and move one half even closer to the center for a smoother transition.
5) Decide whether to keep this Endpoint Adjustment layer as is at 100% opacity, or darken the image by reducing opacity. I chose 85%.
6) Turn to the question of color balance. Within the New Color Boost subgroup are two layers, one each for the A and B. By default both are set at 50% opacity, as is the subgroup itself; these settings put slightly too much color into typical images but allow space for adjustments in many directions. This action is biased toward adding yellow to light areas, so we may not need to adjust the A and B at all. I increased the For Greater B layer from 50% to 60% opacity, thus emphasizing all yellows and blues, and pushing the face toward a golden color.
7) Now comes the question of whether the thing is currently too colorful overall. According to me, it is, so I reduce the opacity of the New Color Boost subgroup from 50% to 35%.
The result is posted as #1029. You can decide whether, as Ken predicts, it's better than either par. I point out that, even with the adjustment toward yellow at Step 6) above, this version still has A=B+3, whereas the par is A=B-2. Those five points of difference mean that this version is rosier, less yellow, than the par. I find that preferable, as noted in my comments to #1006.
1019 (69L 35a 35b, the third most colorful skin) An example of excellent philosophy and dubious execution. The person writes,
The problem is that adding contrast to the face is exactly the kind of "tampering" with the grime that he says he doesn't want to do. It becomes more pronounced than in the original. So I see this one as being about the dirt on the face rather than the face itself.
Given what we have, how do we remedy the situation? For sure, we don't want to wipe the grime out entirely as was done in #1013. As a matter of fact, though, I prefer to reduce this type of garbage by first creating a completely smooth version (usually pretty easy to do) and then blending between us. But why bother to make a substitute for #1013, when the real #1013 is already available to us?
Ingredients: #1013 and #1019
1) Start with #1019. Paste #1013 on top on a new layer.
2) Add a layer mask; to it apply the green channel. This restores all the deep reds, and around half of the face, so the grime is still there.
3) Set foreground color to white. Activate the brush tool. Go into the layer mask and paint over all face areas where dirt is visible. The objective is nearly perfectly smooth skin. No need to be very careful since we won't see this move at full strength.
4) Since #1013 has annoying redness in the cheeks, change layer mode to Luminosity.
5) Reduce the opacity of this layer to restore some, but not all, of the dirt etc. I chose 50% for the sake of argument.
To my mind, reducing the griminess greatly improves #1019 and might satisfy those looking for a less brilliant presentation of the red clothing. The result is posted as #1030. Another comment about this version is found at #1022.
1020 (67L 24a 40b, more than twice as yellow as the next yellowest version) Bill Iverson:
What he likes is for people to check a variety of points before assuming that color is OK. In #1004 I showed how examining several items should have convinced us of a damaging magenta cast. Reading those same points here demonstrates a serious yellowness.
As I’ve commented a couple of times, my preference is for something less yellow, more magenta, than the two pars. Something like #1011, which is more yellow than the pars, can be acceptable if otherwise OK. But #1020 goes too far.
1021 (68L 26a 29b) Nice color, but the face is a disaster area due to oversharpening. I find the clothing oversharpened too, but could live with it. I previously remarked that it's easy to limit the sharpening to one area or another by Blend If. For this, I'll use #1101, which is very similar in color to #1021 but not nearly as sharp. I will now attempt to combine the best features of both.
Ingredients: #1001 and #1021.
1) Start with #1021. Add two copies of #1001 as layers; make the top one invisible.
2) Convert to LAB without flattening.
3) The middle layer now takes precedence, so we see all of #1001 and none of #1021. Go to Blend If, Underlying Layer, and exclude everything that's quite A-positive. This restores the reds of #1021 without its face.
4) Also exclude everything that's significantly B-negative. This restores the blues and purples of #1021 and still doesn't touch the face of #1001.
5) Reduce the opacity of this layer to get a certain amount of detail back into the face. Don't be afraid to go too far, since we're about to get another bite at the apple. I chose 60%.
6) Decide whether to change layer mode from Normal to Luminosity. I can't see much of a difference, so left it alone.
7) Activate the top layer, an untouched copy of #1021. This allows us to soften what's underneath, if desired. I set this layer at 25% opacity, and again found no significant difference between Normal and Luminosity mode.
The result is #1031. Because the sharpening has now been put more where it belongs, I find this version better than either of its parents, and competitive with the other pars.
1022 (66L 33a 33b) Chosen for the par version. At first this one seems odd because of the unusual treatment of making the reds so deep. This has the impact, though, of directing attention to the face. It tones down our perception of the dirt, and makes the color variation in the skin more noticeable.
The person got this effect with luminosity blending, RGB into the red followed by a big contrast-enhancing move in the green. That presumably both darkened and desaturated somewhat the clothing. So, when he went into LAB, he ran the Skin Desaturation action, which enabled him to force a lot of color into the reds later without making the face look ridiculous.
This is one of the better faces among the entrants, but what about these deep reds? Ask yourselves, what are we trying to do here? The instructions for this exercise state that this culture is known for colorful clothing, and the high quality of the fabric. This moves me away from a treatment like #1019, where the clothing is duller. And we don't want #1016 either, because it has no detail in the reds.
#1022 doesn't have either problem. There's certainly enough detail in the fabric. And if the idea to be conveyed is that the fabric is colorful, #1022 makes that point just as well as #1016 does, albeit in a very different way.
Another useful move: this person noted that the background was quite green, and desaturated it--somewhat. I agree with his decision to leave it slightly green, it helps contrast with the face and the darker reds.
It is interesting to compare this one with #1008 and #1015, the other two where darker, deeper reds were used to accentuate the face. You can try blending each with another to see what parts of that strategy work better.
1023 (64L 32a 29b) This person, like me, worked in CMYK because he wanted to sharpen the black. Unfortunately this amounted to locking the barn door after poisoning the horse. Instead of making some use of the red channel for blending, as most of us did, he used the green, thinking thereby to reduce the lightness difference between face and garment. For the face that seems to have been effective, particularly in conjunction with his move of dodging the whites of the eyes. The face of the face is therefore considerably better than the either of the first two pars.
Unfortunately, the green channel was almost solid in the red parts of the garments, so in #1023 those areas are practically flat.
The solution would have been to during the blend of the green, use the Apply Image dialog to mask the green channel with itself, this would have prevented the dark reds from changing.
Another mention of this version is at #1024.
1024 (67L 18a 21b, the third least colorful skin) This particular pot was accused by the kettle of #1002 of being black. And it is true, as the numbers indicate, that this one could use more color. It could also use a better red channel, and particularly it could use the move or dodging the eyes discussed in #1023.
As for the incoming fire from #1002, I strongly prefer this one, not just because it controls the dirt better, but because of the superior handling of the garments.
1025 (67L 25a 27b) Steve Jenkins:
These comments are well taken, but probably most would prefer a bit more vividness than found here.
1026 The par version (67L 28a 30b)
1027 The alternate par (65L 24a 25b)
1028 (71L 33a 35b) Alternate par 2, based on the demonstration at #1016.
1029 (66L 33a 30b) Alternate par 3, based on the demonstration at #1018.
1030 (68L 35a 35b) Alternate par 4, based on the demonstration at #1019.
1031 (66L 24a 28b) Alternate par 5, based on the demonstration at #1021.
Perhaps this deserves a separate thread, but I'd like to drop this in here as this exercises wind up. What interests me most here is the question of the wisdom of crowds vs the judgement of experts, going back the MIT "retoucher" used as a model for the AI programing. For a crowd to be make good assessments, it needs to large, diverse, and not self-conferring. Can a crowd make good esthetic assessments? What happens when the MIT researchers anoint the chosen one to train the AI? How does that bake in bias, and at the same time open up room for expression by creating a norm from to counter? Professionally, I have to land a picture in a spot where the photographer, the client, the model's agency, the ad agency, the stylist, and hair+makeup are all okay with the picture, and not just do that, but satisfy all these stakeholders while doing the thing I'm also hired to do, not have it look like it came out of a major shop, ie, simply checking those boxes and being done with it. It was easy for me to produce images that deviated meaningfully from my anticipated typical submission, yet if was hard for me to produce a par that had a major variance from Dan's, which for me opens more questions in terms of bias/background, set size, and the generalizability of pictorial esthetics.
First of all, most of these versions are quite good. They've taken an excellent original and made it better. A dozen or so are worthy of being chosen for par. So many of the comments here may sound like nitpicking, which I suppose they are in comparison to past exercises where we had a lot of poor submissions.
The notes that accompanied the submissions indicated that people gave a lot of thought to what they were going to try to achieve with the photo. Some of these involved technique, which I’ll get to in a bit, but there are a couple of major artistic decisions.
The one we’ve talked about the most is how to deal with the child’s dirty face. With the exception of one person who decided to retouch out all hint of imperfection, we all acknowledge that this griminess is an important part of the picture, just as, for example, the foreground garments are.
When we identify something as being important the natural reaction is to enhance it. And indeed, the viewer can’t get a full appreciation of the colorful shawl unless we do something to bring out its detail, as all the good versions did.
Dirt on the face of the subject of a portrait is so unusual, OTOH, that no enhancement is needed. We fully appreciate what it represents even in the flat original. Enhancing it does nothing for our understanding of our image, makes it uglier, and makes the picture more about the dirt than about the girl.
As a rule, we don’t like to apply sharpening to the skin of a child, or of a woman of any age. Yes, we sharpen eyes, but not the skin, for fear of emphasizing imperfections.
To see why, compare #1001 and #1007. I choose these two because they are very similar in color. The principle difference is in perceived sharpness; #1007 has more. Rather than inquiring which version is better overall, please split your opinion in two: which is better with respect to the foreground clothing? With respect to the skin (exclusive of eyes)?
For me, it’s an easy answer: I prefer #1007 for the fabric and #1001 for the skin. All that extra definition in the blemishes in #1007 is counterproductive.
Some people did not sharpen the skin explicitly, but applied contrast-enhancing curves that brought the smudges out in the same way. I think that should be avoided. At least one person agreed and did some mild retouching to diminish, but by no means eliminate, the dirt. Makes sense to me.
The other artistic question brings up the old warhorse, when everything is colorful then nothing is colorful. Ken Harris states it thusly:
So again, here we've got expression, situation, and light. For expression, you just need to keep the greatest attention on the face, meaning good contrast, and not over contrasting the rest of it. For situation, that's the clothing, Dan will have much to say about that. I think a defect in many is hitting the knit under hat too hard, which, being close to the face, pulls away from the eyes. OTOH, having detail farther away is less of problem in that the eye can circle around and come back to the center. For light, the image starts with a wonderful blueness to the highlight on the nose. That's a feature.
In other words, should the headgear be almost as colorful as the shawl? Some people seem to think so, but the better versions, like Ken’s, say no. A duller top of the picture makes us appreciate not only the face but the brighter reds beneath it. Often the H-K action, or equivalent, was used to accomplish this.
It is true that the brilliance of the foreground can distract from the face, but it isn’t like it can be desaturated. Some people cleverly made it a deeper, darker red. Meanwhile we all had to be careful to get more shape into the face—some versions had it very flat, almost without a nose.
All these things are controlled by the red channel, which is where serious detail has to be if a red object is to have shape. Normally if we want a shapelier face we just blend the green channel into the red in Luminosity mode and pick an appropriate opacity. Normal pictures, however, don’t have brilliant large red objects. Blending either the green or the blue into the red may help the face but it will be a disaster for the red clothing, which is nearly solid in both green and blue.
All the best versions, needing the extra detail in the red channel, found ways of finessing this problem. Some blended with the RGB composite as a source, some used masking to prevent the clothing from getting too dark, some used more esoteric means like the Channel Mixer.
Several people tried too hard to find reliable points on which to correct the color. Complicating matters, we don’t know enough about regional skintone for people of this ancestry to make tight generalizations. But when we can’t find colors we’re sure are right, we look for ones we’re sure are wrong. A couple of people missed those and got bad color as a result.
Also, Ken noted above the blue lighting, and you may be scratching your head as to where it came from. I know where, approximately, this photo was taken, because the following day Vincent was at the incredible ruined mountain redoubt of Macchu Picchu. You almost got one of his pictures from there to work on instead of Mount Shasta, because he ran into the same sad problem: two days in one of the greatest archeological attractions on the planet, and nary a ray of sunshine.
I got much better weather there on my own visit 40 years ago, and furthermore almost nobody had heard of Macchu Picchu and it wasn’t the Disneyland that it has since become. I had read and ignored many warnings about altitude sickness, because the imperial city of Cuzco, where most travelers stay, is at almost 3,500m/12,000 feet, and much of the surrounding terrain is even higher.
Being slightly younger and fitter than I am today, I was unaffected, until I decided to sample one of the local specialties, a delicious dark lager by the name of Cusqueño. The water used to brew it comes from one of the half-dozen nearby peaks that are 6,000m/20,000 feet or higher. I highly recommend that beer, but not to drink at anything like that altitude. Having made that mistake, I was up all night swilling coca tea and smelling eucalyptus leaves that the locals were breaking under my nose.
I tell this story to suggest that if the air is that thin, you can’t expect the same sort of tolerance for alcohol that you would at sea level, and you can’t expect the same lighting either. That blueness is natural and, as Ken says, it can be exploited. For example, some have noted a blue reflection in the eyes. That means that the reflection is going to be more pronounced in the blue channel than in the red or green. I think it’s a no-brainer to start off by selecting the eyes, then blending the blue into the RGB in Lighten mode on a luminosity layer. It makes the child seem more alive.
More coming in due course.