Red rose: comments on individual versions

Dan Margulis

Not much need to discuss specific techniques here as most people used (with some variations, naturally) the same general approach of attacking the red channel, plus various tweaks with channel curves.  Differences in the individual versions tended, therefore, not to be the result of basic strategy.

There are more than five excellent versions, so please don’t take the selections for par very seriously, I do that quickly and sometimes change my mind after the images are posted.

This posting has a special bonus for those wishing to discuss how red to make the flower. I’ve selected the whole thing in each version and then taken an LAB average of it. Roughly speaking the L value tells how dark the flower is, the A in this particular image defines how colorful it is, and the B/A ratio, which defines hue. This is not a definitive tool but I was interested in what it would show and you may be also. It has the advantage that we can compare flowers that measure similarly to see how, say, different backgrounds contribute.

Explaining: the higher the L value, the lighter the flower (although this can be influenced by an exaggeration of its darker areas). Our versions range from 55L to 30L. The par version is 43L. I would say that anything 50L and lighter appears washed out. Anything 40L and darker is somewhat eccentric although a couple of versions are pretty good.

The higher the A value, the more colorful the flower. It would be difficult to have an average higher than 70A in sRGB without going out of gamut somewhere. Our versions range from 39A to 70A. The par version is 62A. I find that most but not all of the good versions have an average of at least 60A.

The higher the B/A ratio, the more orange the reproduction. A value of 1.0, which nobody got to, would be fire-engine red. Values of .5 and lower technically make the flower more magenta than it is red.  Our versions range from .44 to .81. The par version is .63. This is a fairly small range, indicating that we all basically agree on the hue. Personally, I don’t care particularly for the versions less than .55 although I don’t think it’s disqualifying. I like the .81 version but I surely wouldn’t go farther toward orange.

So, here we go. Remember that when I discuss these values, when I call something the “darkest” or “the most colorful” I am referring only to the flower and the background is not included.


1201 (45L 62A .61 B/A) Chosen for the par version.  Apparently chaos led to a good result. Here is this person’s full description of what he did:

I found it quite easy to create all sorts of monstrosities, but quite difficult to get an image of the flower that looked interesting and realistic. It most often ended up looking quite plastic, or like an artichoke that had been painted red.

I started from the flat version jpg, and made a slight adjustment to the colour (as it was a little cold).

Then the usual shenanigans of making multiple versions with blends, hammers, curves. I also tried some of the techniques listed in the carnival exercise.

Then blended these together to get as best a version as I could.

1202 (42L 54A .59 B/A) This person took the straightforward path to get excellent detailing. Understanding that the red channel is key to contrast in reds, and that it is probably too flat in the original, he blended the green into it, correctly doing so in Darken mode to avoid lightening the leaves. He then applied contrast-enhancing curves to the resulting red and to the green, and changed layer mode to Luminosity. He later doubled down with things like the Velvet Hammer and a false profile routine, but the heavy lifting was done by this luminosity blending.

Nevertheless this version shows a defect. The lighter parts of the flower are too white and the darker parts too gray. They affect our perception of redness and explain the low A value.

One solution would be to make a quick version that doesn’t have the defect and then blend the two. Maybe better: go into LAB and select the rose. On a duplicate layer, run Filter: Blur>Average, This average value isn’t red enough, so you have to run Color Boost or similar curves on top of it. When there is a nice bright red in this average, change the layer mode to Color and reduce opacity to taste. This will affect the lights and darks much more than the rest of the flower. So, at a low enough opacity, it will seem only to correct the defect without really changing the overall appearance.

1203 (36L 57A .53 B/A) This person had never seen #1202 but his opening comment almost seems to have been made with it in mind. He says he Google-image searched for a red rose with white lips, or even very light ones, finding nothing. (He did find that the reverse was not uncommon). So, he writes, 

My first objective (obsession?) was dealing with the too whitish/lightish petal lips -- from lighting, surely, and not true coloration. My early tries, via rgb blending, ended too close to a zebra rose -- yuck.

His solution was tried by many others: a darker rose supports more color than a lighter one. The treatment is a little clumsy IMHO compared to, say, #1213 which is even darker, but gets more excitement into the flower. Here, the person should have tried Auto Tone, which would have revealed the need to add contrast somehow, even if it was nothing more than adding lightness to some of the irrelevant detail in the background.

1204 (52L 57A .46 B/A) This person did the channel blending to get the necessary shape, but was looking for a lighter flower than the others seen so far. He didn’t come up with a “zebra rose” like #1202 because the darkest parts of the flower are still quite red. The whites, however, are quite distracting. Auto Tone would have shown that this version needs improvement.

1205 (53L 70A .45 B/A, the most colorful version of all) This person struggled through five different versions. None of them, however, had the key channel blending, so the red channel here is basically blank and the rose lacks shape. He did try to put contrast in the flower with curves to the L of LAB, but this doesn’t add the needed color variation. 

1206 (49L 61A .49 B/A) It would be interesting to ask for a vote between this version and #1205. This one has better detailing in the flower, but #1205 has a more natural transition between reds and whites. It’s a hard choice, but the solution would be to put the red channel from #1206 into #1205. That would result in something better than either parent.

1207 (53L 52A .65 B/A) This is slightly yellower than others (note the relatively high B/A) but it was a deliberate choice to try to make the overall appearance sunnier. Those who have seen this type of red object many times know instinctively that they must do some blending into the red channel. And some of them realize that H-K can be helpful as well. But this person doesn’t seem to have had such experience. He worked out what to do by himself, but it’s useful to listen to his initial reasoning:

This I found to be difficult as I wanted to get more contrast into the petals without losing credibility in color. And at the same time I wanted more color in the rose.

So, after several false starts, he hit on blending the blue channel into the red at 20% opacity. He then duplicated the result three times, to each of which he tried a different color move, and one of which involved H-K which he described as successful.

The bottom line: a successful version. And next time it won’t take nearly as long.

1208 (51L 65A .55 B/A) This person got attractive color but although he did try out H-K he does not appear to have done any channel blending. As a result, the red channel is largely blank in the center of the flower and overall the rose lacks detail.

1209 (45L 65A .62 B/A) An opening salvo:
More continued thanks for these case studies, I think we're all feeling a bit
case study fatigued, but it was more than well worth it for me, and I hope it was
well rewarding for you, too.

Probably not as much for me as for others, but every now and then a new idea takes shape. This picture is like Seated in the Grass: both feature a large red object in the center, and a green background. In that exercise a number of people, including me, felt that the background should be made unrealistically colorful, as this made the overall effect more pleasing. From which I would assume that the same might be true here. #1220 is the ultimate test of whether this is true, but the person who did #1209 says yes:

Making the greens look a bit too happy was a conscious choice, first to contrast with the reds of rhe flower (hello Mr. Chevreul), but also the image just looked too colorless with the original grayish leaves.

Was he right? These LAB numbers let us find out. They’re almost identical to those of both #1210 and #1214, both of which, especially #1214, have duller, darker leaves. In comparing the three, the rose itself won’t make much of a difference, since they all measure the same—it’s a pure question of background. My vote is in the order of darkness: I prefer #1214, #1210, and #1209 in that order.

1210 (45L 65A .60 B/A) Here is my version from 2017, trying to limit myself to three minutes. It seems to hold up pretty well. I suppose that filling the lighter parts of the flower with a purple/magenta that doesn’t really match the hue of the rest of the flower is better than leaving them neutral.

1211 (50L 63A .44 B/A, tied for the most magenta version) This one may move the leaves too far in the opposite direction, but overall it’s pleasing, and included a blurring routine to try to fill in some of the offensive whiter areas of the flower.

1212 (36L 54A .44 B/A, tied for the most magenta version) A dramatic treatment, intentionally dark produced with a lot of PPW tricks to create contrast in the flower, but derailed by too little saturation in the dark lower part of the flower and the whitish upper part, just as was the case with #1202. Note the low A value in both. This version is best compared to #1218, which is equivalently dark but with far more attention to color than to detail. Blending 50% of that into this one would improve things. Vice versa, possibly.

1213 (30L 52A .77 B/A, the darkest of all versions and the second yellowest) Thomas Hurd has presented his workflow in the main thread. He blames the white areas in the rose that troubled so many of us on a harsh flash. He responded by changing the Temperature in Camera Raw to 5500 and reducing exposure. Rather than blend the green into the red as so many of us did, he blended it into the L of LAB. He used a false profile, H-K, the Bigger Hammer action, and a stiff dose of MMMCB, with an inverted A mask used to prevent the greens from getting out of hand. I find the final result hard to believe because the background is so dark and the lighting of the rose so pronounced, but I suppose this is one way to prevent appearance of those white blotches.

1214 (46L 64A .61 B/A) Chosen for the par version. This one avoids many of the slight problems found in others. The flower is full of life, the background neither too gray nor too obtrusive (see note to #1209 comparing the two, plus #1210, all three of which have nearly identical average values for the flower.)

1215 (55L 54A .56 B/A, the lightest version of all) Seems to be confused as to whether the flower is pink or red. White areas are too pronounced. But several steps created good contrast in the flower.

1216 (43L 53A .58 B/A) Chosen for the par version. If I understand the notes correctly Robin Mark D’Rozario eliminated the white areas by burning the red channel. The rest of his moves to create the flower are familiar and are described in his note to the main thread. He decided to move slightly away from a fiery red and toward pinkness. Also, he found that the green background was too light and detracting from the flower, so he made a darker version and blended it into the background at a low opacity. The question is whether the flower should have been redder, see #1219, whose flower is the same hue but darker and more colorful.

1217 (51L 39A .74 B/A. the least colorful of all versions) The leaves don’t have to be a bright green but I don’t think they should be gray either. The idea of putting a purplish tint in the white areas is similar to that of #1210. But comparing the two we see that this version, though almost the same in the lighter areas, could have used a deeper, darker red at the bottom of the flower.

1218 (45L 67A .81 B/A, the yellowest of all versions and the second most colorful) Chosen for the par version. Paco Márquez, who posted a description of his workflow to the main thread, decided that most of the hard work would be done on the flower, with the background excluded. I understand that there is something to be said for a soft-looking rose but I felt that it could use a bit more detail. 

Paco has now posted that he has reached the same conclusion himself, and he would now have blended some of the par version into his.  Personally, in situations like this I prefer to blend with things that emphatically don’t have the problem. The par version is much better on the whole than #1212, but since #1212 takes contrast to an extreme I’d consider it a better choice for this blend.

1219 (37L 59A .58 B/A) Chosen for the par version. Here’s mine, with the detailing mostly due to curves, as well as blending and H-K. I wanted to de-emphasize the white areas and wasn’t happy with my 2017 version (#1210) because I had filled them with a different hue.

To get a somewhat better result this time, on a duplicate layer I made a very rough selection of most of the flower (including all the lighter areas) and ran Filter: Blur>Average, immediately fading it to Darker Color so that nothing would get lighter. I added a mask based on the L channel and edited it sharply so that it allowed darkening only in the lightest areas, then reduced layer opacity to taste. This method has the advantage that nobody can claim that I used the wrong color to fill the holes.

For those wavering on the question of how red to go with the flower, it is useful to compare this one to #1216. Both dispense with the white areas in the flower, the green backgrounds are very similar, and the flowers are equally dark. The major difference, then, is the extra saturation (higher A) in the rose of #1219.

1220 (46L 57A .51 B/A) An fine flower, produced by an experienced person in five minutes. The background is jarring not just because of its saturation but because of its hue. Measuring a point in the large leaf to the right of the flower yields  50L(38)a24b as opposed to 34L(18)a21b for the same point in the par version. I normally recommend that the absolute value (distance from 0) in the B channel of greens should be at least 1.5x that of the A. The par channel doesn’t comply, but rose leaves, being quite dull and blue, might reasonably fall outside of this recommended range. This version, however, has the A channel twice as negative as even that relatively extreme measurement. So, the leaves are too blue/teal.

Even without the hue issue, however, there is the question of whether to make the leaves bright. #1209 does that more convincingly. They are correctly more yellow, but the guy admits he deliberately made them “a bit too happy.” Personally, I don’t think it works here even if a similar move did work well in the Seated in the Grass study. I commented on why this might be in the main thread.

1221 (54L 60A .48 B/A) Appropriately enough, we close with the average of the five MIT retouchers. As it is better than any of their individual versions, it’s the only one shown here.

1222 (43L 62A .63 B/A) The par version.