Veiled Bride: Blending, and Eyes


Dan Margulis
 

Here’s a little exercise that you can try while I prepare a few more posts about this veiled image.

I hope you will agree with me that #319 is hideous, way overdone, she looks like some kind of vampire.

Download it anyway. Blend it into your own version, 25% opacity, Normal mode.

You would think that blending with something that ugly into is pretty stupid. But I’ve just done the exercise with all our images, except #308, where it was used already.

Results: this blend didn’t seem to help #307 very much if at all. Every other image was clearly improved. In many cases it would have been even better at a higher opacity.

Food for thought, no?

Dan


Gerald Bakker
 

On Wed, Jun 3, 2020 at 03:31 PM, Dan Margulis wrote:
You would think that blending with something that ugly into is pretty stupid. But I’ve just done the exercise with all our images, except #308, where it was used already.

Results: this blend didn’t seem to help #307 very much if at all. Every other image was clearly improved. In many cases it would have been even better at a higher opacity.

Food for thought, no?
Absolutely. I tried this on my own version, and it's an improvement indeed, although (as expected) not in every respect. The face gets better, the veil also except for the very top, and the background except for the greens and flowers which I find getting too colorful. Of course, these areas can easily be masked out, leading to an even better version.

If this is true for almost every other version, well, that's stupifying. The question is, what makes this #319 so special? Is it just one of the group members' submissions? Or was is constructed in some particular way to make it behave like this? I really wonder what conclusion we can draw from this experiment.  
--
Gerald Bakker
http://geraldbakker.nl


Dan Margulis
 



On Jun 3, 2020, at 2:58 PM, Gerald Bakker <gc.bakker@...> wrote:

If this is true for almost every other version, well, that's stupifying. The question is, what makes this #319 so special? Is it just one of the group members' submissions? Or was is constructed in some particular way to make it behave like this? I really wonder what conclusion we can draw from this experiment.  

Well, stay tuned. I wanted to throw out the teaser so that people will realize this is not some kind of parlor trick but a legitimate means of improving an image. Details to follow.

Dan


Harvey Nagai
 

Interesting...

The immediate thing that struck me about 319 was the excessive local contrast/clarity.  My tactful opinion is that it is the antithesis to most peoples' notions of a wedding portrait.

But there staring out of it is... that light blue eye, which (I acknowledge) would improve any rendition that didn't lighten the eye enough.

Mash that clarity with a double dose of ACR's texture slider leaves something that LOOKS more blendable. The broader facial tones (if not colors) are not that bad.

Yes?

No?


Dan Margulis
 


On Jun 3, 2020, at 2:58 PM, Gerald Bakker <gc.bakker@...> wrote:

The question is, what makes this #319 so special? Is it just one of the group members' submissions? Or was is constructed in some particular way to make it behave like this? 

Something quite like that, yes. It is obviously not intended to be seen by itself; I have an “official” version that I am working on, and am more or less satisfied with. How to improve it? I ask myself,

*Should the eyes be made lighter?

*Should there be more shape in the facial structure, the cheeks, nose, etc.?

*Should the MMM effect be further exaggerated, to get more variation in the skintone?

*Should the veil get lighter, while the skin stays the same or even darkens?

I need not be a mind-reader to suppose that almost everyone else is or should be asking themselves the same questions. Wherefore if I find a solution to my problem it is likely to be a partial solution to everyone else’s as well.

That solution is #319, a version that is otherwise OK but attacks all four of the above in exaggerated form. The plan is to do a series of blends to adjust each factor to taste. When doing a version specifically for blending, it pays to go for overkill. Obviously I do not wish the eyes to be as light as in #319, but this is why we have an opacity slider.

As a rule it is easier to put in too much color and then back off than it is to try to force sufficient color into an overly dull image. That is the theory of the Color Boost acttion and it works here as well. Without this strategy it is harder to find the perfect balance of the four factors above. Since most people will stay on the conservative side it is to be expected that a small helping of #319 will help in most cases, although a more careful kind of blending with it would be even better.

I made #319 because I already had a fairly good version, namely the one I made in 2017, #322. But when I was working on the MIT set back then, I was trying to hold myself to three minutes per image, so there was no time for adventures like a separate version for blending. For our case study it became possible. Nothing to lose, because if #319 was useless I’d just trash it and submit #322 as is.

Making #319 was easy. In the next post I’ll give a general recipe for lightening blue eyes.

Dan


Dan Margulis
 

Some people had trouble lightening the blue eyes in this image, particularly since one eye is partially covered by the veil. There’s an easy way to do it with channel blending.

In any face, with rare exceptions, the channel lineup from lightest to darkest is: R-G-B.

Two exceptions:

If the subject is wearing lipstick, the lineup is usually R-B-G.

If the subject has blue eyes, it’s B-G-R.

This opens easy blending possibilities if the target is either the lips or the eyes.

If the lips are too dark, on a duplicate layer blend the blue into the green, Lighten mode. It may be necessary to isolate the face. It may or may not be necessary to change mode to Luminosity.

If the blue eyes are too dark, on a duplicate layer set to Luminosity mode blend the blue into the red, Lighten mode. If there is anything else blue in the picture, a very rough selection around the eyes will be sufficient for the blend, since nothing else in a face is going to be blue.

In the veiled bride case, to lighten the eyes I needed no selection at all, since there are no blue objects other than the eyes anywhere.

To make #319 required an extreme move, since I wanted the eyes to become grossly too light. I also wanted to increase shape in the cheeks/nose and lighten the veil. So it becomes slightly more complicated. I’ve uploaded six of the step results to our folder, https://groups.io/g/colortheory/album?id=247497. There are many variations possible but they all depend on creating an alpha channel that is a grossly lightened copy of the blue.

When I did this as a live job, I started from scratch, but for this demonstration I will start with #322, which is my 2017 three-minute correction.

*The original blue channel (#322c) is full of contrast in the face, albeit overly dark, and also has relatively light eyes.

*Copy the blue into an alpha channel and drastically lighten the quarter tone. The resulting alpha channel is #322d,

*On a duplicate layer, apply the alpha channel to the RGB composite, Lighten mode, opacity (say) 50%. Then change the layer mode to Luminosity. This lightens the eyes in both the red and green channels, and since the green is the most powerful channel, the effect is noticeable. There is modest improvement in the face structure as well. We are now at #322a—compare it to #322.

*Add a new composite layer. Apply the alpha channel to the red, Normal mode, 100% opacity. This further lightens the eyes in the red but adds a great deal of depth to the face.

*Change layer mode to Luminosity. The result is #322b. The original red channel is #322c; the red channel after all these shenanigans is #322d, which explains why #322b has such light eyes and strong detail.

*Not shown: with the desired contrast now established, I would start jacking up the color with, perhaps MMM, and Hue/Saturation aimed at the blues. That would get me something similar to #319 which could now be used to blend into any other version.

Dan


ROBIN MARK D'ROZARIO
 

Thanks very much this. I need  to practice more with channels for contrast,would save  me a whole lot of time and needless experimentation with  dead end versions.