King of Beasts: Dan's comments


Dan Margulis
 
Edited

This exercise is the equivalent of the Monument Valley study of last year, in the sense that it’s the one image out of the ten that most clearly shows the advantages of PPW. Indeed, this is very like the canyon images that are commonly used to promote work in LAB: lots ofsubtle colors that need to get more variation, yet we can’t afford to make them too lurid.
 
There are not as many traps to fall into as in our last three case studies. Furthermore, the highlight and shadow points are obvious. For all these reasons, I consider this the easiest exercise of the ten, with the possible exception of Hotel Lobby. Like that one, here there are many blending possibilities but one (green into red, Luminosity mode) stands out and was adopted by many of us. Also, the MMM script, which is crucial for getting variation into the lion, was generally used. For these reasons we have a dozen or more decent entries that could reasonably be chosen for the par.
 
We expect, also, that we should do much better than the MIT retouchers, who presumably used none of these tools. And this is indeed the case. Their average result happens to be #601. Most of us did significantly better, and better also than their two best individuals, who are #623 and #631.
 
Yet a lot of our work that is obviously better than #601 doesn’t stand up so well next to the par. There is no disgrace is preferring the par to your own version; in our studies typically only one or two people, if that many, can honestly say they consider their version to be as good as or (occasionally) better than the par. And if we do have a slight preference for the par, it’s usually because the par tends to have more accurate color, not so much better contrast.
 
If you’re mystified as to why the par is better, here’s a valuable test. Put the par as a layer on top of your version. Toggle it on and off. Now, change its mode to Luminosity, so that what you see is your color, with the par’s contrast. Again, toggle between the two. Then, change the layer mode to Color, so that you see your contrast, its color. And toggle again.
 
The normal result is that you will decide that the par is slightly better in both respects. Occasionally, though, you will find that the two versions are about equal in one respect, but in the other, the par is grossly superior. We saw and discussed an example of that in the Mantillas exercise where one person actually had better color than the par’s, yet his contrast was much worse. I regret to say there are around half a dozen such instances in this set. That shouldn’t be happening.
 
If the par’s luminosity is much better than yours the explanation is probably simple. You probably didn’t do the appropriate channel blending.
 
If the par’s color is much better than yours the explanation is likely more complicated, and in a way that is applicable to other similar images. Start by asking yourself: what is the basic color of this animal? I think we can agree that it is neither lemon-yellow nor tomato-red, but something between those two extremes. Which of these is the best description?
*Reddish yellow
*Yellowish brown
*Brown
 
What are the desired values, then, for a typical patch of darker fur, nearly neutral, in the top half of the animal’s flank?
 
Brown is a species of red, although much less saturated than a tomato. If we measure LAB values, in principle we should get something like A=B with both positive. That, however, can only be quite a dark brown, such as a walnut table, or a dark brown horse. As the brown gets lighter, it also needs a higher B value. So, a light wood like beech, or a lion for that matter, can have the B twice as high as the A. Going any further than that technically makes it a reddish yellow rather than a yellowish brown.
 
What happens if you ignore the need for a higher B, and have A=B? Well, then you get something like #625. The hue is perfectly fine for a dark brown animal, but it doesn’t work in a lion. Or, in another way, you get #614, which has a much better body. The mane admittedly needs to be redder than that. But it can’t go all the way to A=B.
 
I have measured a representative point in the body, and also in the part of the mane that we expect to be somewhat redder. What would acceptable values be for each area?
 
We should acknowledge that the lion is being hit by sunlight that could push it more toward yellow. So for the body, I think I’d accept anything up to about B=A*2.5. For the mane, which we know is redder, I’d say that anything worse than B=2A is definitely wrong.
 
I happen to have measured the same points in every version. I also measured the highlight areas, but we didn’t seem to have many problems there, so no need for discussion. With respect to the lion’s body, I have a list of all those who have a B/A ratio of 3/1 or higher. Also, a list of those having the mane at significantly greater than 2/1. In the first category there are way too many guilty parties, namely ##603, 605, 608, 610, 611, 612, 613, 617,  618, 619, 621, 623, 624, 626, and 635. On the second count, the following are indicted: ##601, 603, 621, 623, 629, 630, and 631. And on the other side, the following have it as essentially a pure red: ##605, 614, and 628.
 
What are these people guilty of, if the B is too high? Another way of putting it would be, the A is too low. Four other LAB-based descriptions might apply: too yellow, not magenta enough, not blue enough—or too green. Those last two words, if true, explain why only two of the above-named versions made it even into my top ten, which I picked without looking at any numbers. Maybe you can get away with such slight inaccuracies in a more difficult image like the three previous studies, but where almost everybody is getting something halfway decent the error is noticeable.
 
These images likely fall in that unpleasant category described above, where the par is markedly superior in terms of color. Translation: this is the end of a three-part series in which coolness vs. warmth is a key issue. In Beach at Sunset, those who engineered warmth into the scene were successful. In Bellagio, it was the reverse. And here? You be the judge.
 
Because a problem with these same colors often arises in other contexts, I’ll have a pair of demonstrations on how to rectify it with a simple LAB blend tomorrow when I post comments on individual versions.
 
A couple of lesser points.
 
First, it’s very helpful to lighten the eyes. A lot of us did so and it makes the animal seem more real.
 
As for desaturating the red nose, or any of the greenery, I’d say it’s personal preference. Some of us were willing to let the background have some color, others tried to suppress its saturation, or even make it darker. A big controversy was what to do with the yellow area in the background, Some think it suggests sunniness, even if unrealistic. Some say that lions are designed to blend into the environment. They therefore lessened, or retouched it out. OTOH, this particular lion wasn’t given a choice of background environment. So I am fine leaving that yellow in. It’s a matter of taste, as is leaving the log behind the animal slightly blue. In principle it’s a by-the-numbers violation; in practice, we live in an age of color grading and that blue helps bring out the subtleties in the yellowish-brown (repeat, yellowish-brown) of the lion.
 
Dan

Join colortheory@groups.io to automatically receive all group messages.