Convert to LAB before adjusting the blending options. In this case, the magenta blendif slider can capture the hair piece without touching the faces. Your really have to do this anyway, as you are planning not to flatten until you move to LAB and your blending options will be lost after you convert.
That's an interesting suggestion and I can see why it would have worked better in this case. The Margulis recipe summary on page 324 does not suggest this - in fact it advises the opposite (to set the blendif sliders while in RGB). But, in the several examples he goes through he sort of proves that there really isn't a recipe for this step since he uses a different method on nearly every image. That's another one of the reasons why this isn't really a recipe. You have to fully understand every step so you know how to adapt them to a particular image.
Yes and no. It is exactly the same thing as Apply Image followed by changing the blending mode to luminosity. Once you do this your future options are exactly the same as if you had used Apply Image. No differernce at all. You have not opened any additional options for yourself.
I find the image a lot more self describing when I have a copy of the green channel with a luminosity blend mode than when I have a dupped layer that has somehow been changed (via Apply Image). The former shows me exactly what was done to it - the latter does not. It also shows me the actual green channel that I was blending (in B&W) so I can see why things came out too dark if they did. The Apply Image method doesn't show me any of that after-the-fact. Since either way will work, folks can make their own choice.
There you go again, flattening in RGB instead of LAB. You really want the luminosity blend to take place in LAB so you get the benefit of impossible colors if possible. Also, I find the LAB blendif sliders more versitle as discussed above.
Just curious about a question: Do blendif sliders set on the luminosity layer in RGB mode convert properly to LAB mode? How would they get represented in the LAB mode UI? Would a green blendif setting get converted into pieces of an L, A and B blendif setting in LAB mode?
My PS action basically make the layers you describe but with overlay technique. See my initial post on this thead and search for "action". Also, I actually worked out curves that are identical to the overlay blends. See: http://dgrin.com/showpost.php?p=202259&postcount=71
Edgework has some even more compelling reasons to do this. Read his posts to the portrait thread.
That helps a lot to understand the curves that are the same as overlay mode. Now the obvious question. There's no magic of overlay mode. It just happens to be equivalent to these curves and convenient to use. So, why these specific curves for portraits? Wouldn't the specific curve that works best be quite image specific, just like it is when applying contrast curves to the L channel? And, wouldn't using curves instead of overlay give you a lot more flexibility in this regard?
Also, this begs the question of when should you push in the ends on the A/B curves and when should you use an S-curve and when a combination of the two? Chapter 12 sometimes does one and sometimes the other. Is there any easy way to tell which is more appropriate for a given image?
For an RGB curve in RGB mode, I know to only push in the ends when the histogram is either empty at one end (so pushing in the end just spreads out your tones into unused tonal space) or when I'm purposely pushing some things to full black or full white (more commonly done on a mask than on an image), but sometimes done at the black-end on an image to give it a bit more pop. Should I be thinking of the A/B channels the same way and looking at their histogram? Or does LAB work differently in this regard?
I'm a bit confused because I've always wondered why pushing in the ends on an A/B curve (the basic Chapter 1 technique) doesn't clip something like it does on an RGB curve.
No, sharpen in LAB or perhaps in CMYK if you know what you are doing and why. There are very good reasons for this. Perhaps you actually did this, but it's not what you described.
I haven't fully adopted Margulis' sharpening techniques yet. Not because I don't like them, but I just haven't fully gotten there yet.
Also, this begs the question of when should you push in the ends on the A/B curves and when should you use an S-curve and when a combination of the two? Chapter 12 sometimes does one and sometimes the other. Is there any easy way to tell which is more appropriate for a given image?
For an RGB curve in RGB mode, I know to only push in the ends when the histogram is either empty at one end (so pushing in the end just spreads out your tones into unused tonal space) or when I'm purposely pushing some things to full black or full white (more commonly done on a mask than on an image), but sometimes done at the black-end on an image to give it a bit more pop. Should I be thinking of the A/B channels the same way and looking at their histogram? Or does LAB work differently in this regard?
I'm a bit confused because I've always wondered why pushing in the ends on an A/B curve (the basic Chapter 1 technique) doesn't clip something like it does on an RGB curve.
In truth, Dan doesn't actually suggest using curves instead of the A/B overlay trick, so the fact that the curves that mimick that proceedure look the way they do is somewhat beside the point. The feature that those curves, the channel overlays, and the techniques from chapter one all share is that they are perfectly symmetrical shifts. This keeps things safe, in terms of avoiding gut-wrenching shifts in color where you aren't looking for them.
The LAB color curves are so powerful, and sensitive, you wouldn't want to simply whip up an "S" shape, as would be appropriate in CMYK if you see that magenta is flat. The whole second half of Chapter 12 deals in-depth with non-symmetrical moves, and provides some careful guidelines for identifying when they're appropriate, and how to avoid disaster.
As for why the ends of the A & B curves don't clip, that's because your monitor's RGB gamut has already done all the clipping. The colors out at the edges are outside every gamut imaginable, and probably some that can't even be imagined. However, just as with RGB and CMYK curves, the difference between pulling an anchor point, and establishing an "S" shape has as much to do with the way the move is distributed throughout the rest of the curve, as with the point being shifted. If the lowest value in, say, my cyan channel is 16, and I have an area that I definitely want to go white, I can either set my anchor for Input=16, Output=4, or I can set a point at 16 and pull it down to 4, leaving my anchor at 0. My highlight will reach the desired value either way, buut the rest of the tonal range will be very differently affected, depending on which method I go for, as the very different shapes off the curve suggest. The first version will be darker overall; the second version will not only be lighter, but the lightening increases as the bottom point is approached.
Now, here's the real benefit of this method over Margulis'. Margulis' technique requires you to guess what percentages to use in Apply Image, sometimes using different ones for the A and B blend. If you follow his technique and your first guess isn't right, you can't change it. You have to undo and redo it. If you use the adjustment layers, you just tweak either the A channel, the B channel or both. No undo. No starting over.
My PS action basically make the layers you describe but with overlay technique. See my initial post on this thead and search for "action". Also, I actually worked out curves that are identical to the overlay blends. See: http://dgrin.com/showpost.php?p=202259&postcount=71
One thing to note, using layers over Dan's method: if both A and B layers are at 100%, of course only the top will show. Reducing this to 50% will balance both channel blends, but the intensity will be at 50% for each. That's the max, for a balanced effect. If, as I have done, you place the B layer on top at, say 60%, that leaves your A layer to come in at 40%. As to whether it's a bug or a feature of using layers, I'd weigh in on the side of a feature, since every time I've tried it Dan's way, blasting the channels on top of themselves at 80—100%, I've ended up dropping the resulting layer's opacity to around 40% or thereabouts. However, it's important to recognize that substituting layers for each channel blend is not a precise duplication of Dan's method.
I'm still on the fence about this technique, as opposed to MFM curves, as opposed to simple A/B steepening, as opposed to doing it the old fashioned way in CMYK. I've gotten my best results so far through a combination of overlays and MFM curves, but I've still gone back into RGB or CMYK for some final moves that aren't available in LAB (Luminosity contrast to the magenta curve, selective color to add a little cyan to Reds and Yellows).
Where all this is heading, clearly, is yet another extention of the age-old Photoshop rule: there's always another way to do it. Trying to think in terms of this step, or that step, is shortsighted. It's more a function of this concept or that concept, and picking and choosing as conditions, and knowledge, dictate. As I've become more and more comfortable visualizing what happens when the A and B curves are transformed, I find that I am starting to see all the techniques in their proper perspective: more ways to do things to color.
There are two ways to slide through life: to believe everything or to doubt everything; both save us from thinking.
—Korzybski
That's an interesting suggestion and I can see why it would have worked better in this case. The Margulis recipe summary on page 324 does not suggest this - in fact it advises the opposite (to set the blendif sliders while in RGB).
Let's forget about what Dan says to do. You are right, it's not really a recipe. In some sense, none of what Dan teaches is a recipe; rather it's a way of thinking about this stuff, a general approach, and a bunch of tools to implement it.
But there is more than one reason to flatten in LAB instead of RGB. LAB handles extremely light and extremely dark colors better than RGB or CMYK can because of it ability to express them as impossible colors. It's important to understand this if you are really going to understand the differences between working in LAB and the other color spaces. At this point you can just use dgrin search for impossible colors to find out more.
So, unless you have a reason not to, take the luminosity blend to LAB before flattening it.
I find the image a lot more self describing when I have a copy of the green channel with a luminosity blend mode than when I have a dupped layer that has somehow been changed (via Apply Image).
After Apply-Image in "Normal Mode" at 100% opacity with the composite (RGB) as target and the green channel as source, the result will be EXACTLY THE SAME as if you selected the green channel and pasted it into a new layer. You can take my word for it, or you can do an experiment: Do it both ways into separate layers. Then set the blending mode to "Difference".
That helps a lot to understand the curves that are the same as overlay mode. Now the obvious question. There's no magic of overlay mode. It just happens to be equivalent to these curves and convenient to use. So, why these specific curves for portraits? Wouldn't the specific curve that works best be quite image specific, just like it is when applying contrast curves to the L channel? And, wouldn't using curves instead of overlay give you a lot more flexibility in this regard?
Very good questions. I went through exactly the same reasoning. Edgework, too. He concluded that the MFM curves generally worked better than the Overlays. I've found that the overlays work really well on images with nice color balance to start with and they are pretty easy to accomplish. If there is interesting light (as with the ballet pictures) or the skin tone is too flat (not just too gray), I like the MFM method. Whatever. Really all three things are just variations on a basic simple theme: get extreme color variation in a layer and then lower it's opacity until it looks good.
Also, this begs the question of when should you push in the ends on the A/B curves and when should you use an S-curve and when a combination of the two? Chapter 12 sometimes does one and sometimes the other. Is there any easy way to tell which is more appropriate for a given image?
Unfortunately, not that I know of. With experience and trial and error better results come more easily.
For an RGB curve in RGB mode, I know to only push in the ends when the histogram is either empty at one end (so pushing in the end just spreads out your tones into unused tonal space) or when I'm purposely pushing some things to full black or full white (more commonly done on a mask than on an image), but sometimes done at the black-end on an image to give it a bit more pop. Should I be thinking of the A/B channels the same way and looking at their histogram? Or does LAB work differently in this regard?
Quite differently. The L curve is like the RGB composite, except it is guaranteed not to affect color, only luminosity. Take what you know about the RGB composite curve and apply to the L curve and you'll feel pretty much at home.
I'm a bit confused because I've always wondered why pushing in the ends on an A/B curve (the basic Chapter 1 technique) doesn't clip something like it does on an RGB curve.
Generally the stuff it clips is so far out of gamut that you can't even imagine it. Run your mouse over one of your images with the button depressed and see where the point is on the curve. You won't have anything out in that last 20% on either side.
I haven't fully adopted Margulis' sharpening techniques yet. Not because I don't like them, but I just haven't fully gotten there yet.
Sharpen the L channel. The results are always better than sharpening in RGB. If you really know what you are doing, there are times when you should sharpen in CMYK. See my sharpening tutorials.
Quite differently. The L curve is like the RGB composite, except it is guaranteed not to affect color, only luminosity. Take what you know about the RGB composite curve and apply to the L curve and you'll feel pretty much at home.
After Apply-Image in "Normal Mode" at 100% opacity with the composite (RGB) as target and the green channel as source, the result will be EXACTLY THE SAME as if you selected the green channel and pasted it into a new layer. You can take my word for it, or you can do an experiment: Do it both ways into separate layers. Then set the blending mode to "Difference".
I understand and agree that the resulting image is exactly the same. I was just pointing out that it feels to me like the difference between using Image/Adjustments/Curves on a dupped layer vs. Layer/New Adjustment Layer/Curves. The latter is a lot more obvious what was done after-the-fact than the former, though they achieve the same result. We're just talking about a style choice for how one wants to accomplish the same result.
Very good questions. I went through exactly the same reasoning. Edgework, too. He concluded that the MFM curves generally worked better than the Overlays. I've found that the overlays work really well on images with nice color balance to start with and they are pretty easy to accomplish. If there is interesting light (as with the ballet pictures) or the skin tone is too flat (not just too gray), I like the MFM method. Whatever. Really all three things are just variations on a basic simple theme: get extreme color variation in a layer and then lower it's opacity until it looks good.
Some variations don't require extreme manipulations that you then lower the opacity. In some cases you can just adjust to taste directly. Again, just a style choice for how you want to do it.
Generally the stuff it clips is so far out of gamut that you can't even imagine it. Run your mouse over one of your images with the button depressed and see where the point is on the curve. You won't have anything out in that last 20% on either side.
OK, that explains it. It is clipping, but there's usually nothing there to clip. That restores my sense of understanding how curves work. I sure wish CS2 would display the histogram in the curves dialog like it does in the levels dialog, in the ACR curves dialog and like other competitive products do. That would make this a lot of this easier to see.
Sharpen the L channel. The results are always better than sharpening in RGB. If you really know what you are doing, there are times when you should sharpen in CMYK. See my sharpening tutorials.
I'm getting there. One thing at a time. From what I've read there's not much difference between sharpening the L channel in LAB mode versus sharpening in RGB mode and then fade to luminosity. Also, is there a way to see the full color version of your image while you are sharpening the L channel? Or do you have to only look at the L channel while doing the actual sharpening?
Also, is there a way to see the full color version of your image while you are sharpening the L channel? Or do you have to only look at the L channel while doing the actual sharpening?
Ah. Select the L channel by clicking on it. Image display goes to B&W. Click on the visibility checkbox next to the LAB composite (the eyeball). Now only the L channel is active, but you see all three channels and the image will be in color. The display in the USM dialog is still B&W, unfortunately, but you can see what you are doing with preview on the main image.
Ah. Select the L channel by clicking on it. Image display goes to B&W. Click on the visibility checkbox next to the LAB composite (the eyeball). Now only the L channel is active, but you see all three channels and the image will be in color. The display in the USM dialog is still B&W, unfortunately, but you can see what you are doing with preview on the main image.
Select the L channel and then hit the tilde key. Easy-peasy.
Ah. Select the L channel by clicking on it. Image display goes to B&W. Click on the visibility checkbox next to the LAB composite (the eyeball). Now only the L channel is active, but you see all three channels and the image will be in color. The display in the USM dialog is still B&W, unfortunately, but you can see what you are doing with preview on the main image.
Select the L channel and then hit the tilde key. Easy-peasy.
Ctrl-1 ~
Dan Dill
"It is a magical time. I am reluctant to leave. Yet the shooting becomes more difficult, the path back grows black as it is without this last light. I don't do it anymore unless my husband is with me, as I am still afraid of the dark, smile.
This was truly last light, my legs were tired, my husband could no longer read and was anxious to leave, but the magic and I, we lingered........" Ginger Jones
K, I finally had a chance to try this technique out. On Xmas eve of all times...
Anyway, here goes. I really think I went too far. In fact I KNOW I did. I ran it past Andy, and basically said WTF? So I scaled back on the saturation. I went back and pulled back on the HIRALOAM. I just pulled back overall, because I was too gung-ho. I also had a yellow cast that I wasn't careful enough with early on, and it got worse.
Before/After:
i've just finished work on an image that had a very strong yellow/orange cast. so back to the drawing board. so i was going back over the steps and i used dan's "apply image" instead of cutting and pasting the green channel. well, i got an entirely different result. it took ages for me to work out why i was getting such a different result.
i make a new layer, apply image to the background and then set the resulting layer to luminosity. entirely different result and effect to pasting in the green channel. or am i doing something wrong?
i'll have to work out how to get photos into this thing...
Thanks. I guess I'm trying too hard to avoid *any* area in the skin where the A is more than the B. Maybe I have to relax a little and let parts of the skin be red.
And I'd like to know what you think aside from that...
Thanks. I guess I'm trying too hard to avoid *any* area in the skin where the A is more than the B. Maybe I have to relax a little and let parts of the skin be red.
And I'd like to know what you think aside from that...
no, you don't have to avoid A being more than B. you've done what i did with this image... missed a cast because it makes B more than A throughout the image. so i ended up with a very orange result.
even in a very pale skin, you'd at least expect the cheeks to be tending to the A.
but in my picture, i corrected the cast. that leaves his skin mainly more A than B. but checking margulis again, he says in a dark haired male with some sun exposure, you'd expect this. she's wearing makeup. and she's towards the pinkish side anyway -- a bit of a rose. and both these kids are aussies in summer and have plenty of colour so i brought the A/B blending right back. i'm often having to remove colour. i checked margulis's first example and she's very pallid. so the recipe works very well.
no, you don't have to avoid A being more than B. you've done what i did with this image... missed a cast because it makes B more than A throughout the image. so i ended up with a very orange result.
even in a very pale skin, you'd at least expect the cheeks to be tending to the A.
but in my picture, i corrected the cast. that leaves his skin mainly more A than B. but checking margulis again, he says in a dark haired male with some sun exposure, you'd expect this. she's wearing makeup. and she's towards the pinkish side anyway -- a bit of a rose. and both these kids are aussies in summer and have plenty of colour so i brought the A/B blending right back. i'm often having to remove colour. i checked margulis's first example and she's very pallid. so the recipe works very well.
i can only post one pic...
Oddly enough, what is usually considered LAB's strongest feature, the ability to deal with contrast and color in totally separate operations, can sometimes be a problem with faces. In this picture the problem is not so much with the color (the numbers are certainly within range) but with the lack of contrast. Pulling the highlights in the lightness channel gives an odd result, not at all natural looking.
Good old CMYK is a powerful space to work with faces, since the three color channels are so distinct in the way that they contribute to both the color and shape of a face. In your image, both Magenta and Yellow need to be stretched downward, like so:
Then, just to make sure the overall image didn't go too cold, I added a second curve in Color mode to put back a touch of warmth. A magenta anchor at 64 moved to 66, and a yellow anchor at 64 moved to 68. I also used a selective color layer, also in Color mode, where I set Cyan for -21 in the yellows. This had no effect at all on the skin but it helped keep her hair from pushing to the greenish side.
Trying to attack the subtle shift in the skin tones is asking a bit more from LAB than it's designed to accomplish. Similar moves in the Green and Blue channel get similar results in RGB, but I think CMYK is the space of choice for fine tuning skin tones. As for the fear that color might be lost in the translation, those tones out of gamut are usually the kinds that don't look natural anyway: bright blues, greens and fire engine reds.
There are two ways to slide through life: to believe everything or to doubt everything; both save us from thinking.
—Korzybski
thanks edgework
i kinda don't know what i'm doing here... . i had to put in more blue to correct the original orange cast, the AB blend ripped it out and now you've put some back. why am i doing the AB blend? so i have to go back and see if i can get something like what you've got without all the steps. my head hurts! but thanks.
i kinda don't know what i'm doing here... . i had to put in more blue to correct the original orange cast, the AB blend ripped it out and now you've put some back. why am i doing the AB blend? so i have to go back and see if i can get something like what you've got without all the steps. my head hurts! but thanks.
My experience with the AB blend is that it obliterates the cyan information, which is exactly what you would expect. The A/B channels are pure color information and, in a face, they will emphasize magenta and yellow. The overlay operation simply makes whatever is there that much more so. Since the lightness channel stays the same, the overall value cannot get darker to account for the added magenta and yellow, so guess what suffers? Cyan.
I've still found it to be a really useful step—in fact, the whole recipe is useful—but in context. When I posted a sample earlier in this thread, I followed Rutt's suggetion of placing each channel's overlay operation in a separate image on a separate layer. This restricts the maximum percentage for both channels to 50%, though I set the B layer on top at 60%, giving it the edge over A. I also put both layers in a layer set that I reduced in opacity as well. A full intensity channel overlay blend in both A and B is something that I have not found to be workable, and there are other ways to achieve the same effects, with more options and subtlety in the moves. Keep in mind that these recipes are things that Dan Margulis evolved for people who want to derive some benefit from the power of LAB but who aren't really interested in getting deep into the dynamics of what is happening. A one-size-fits-all approach will necessarily compromise across the board in exchange for delivering a more-or-less decent result.
I've also found it useful, after all the LAB work is finished to take the image back into RGB or CMYK and use Selective Color on the Reds to add a touch of cyan, just to cool things off. One way or another, you will need to compensate for the drop in cyan information, however.
There are two ways to slide through life: to believe everything or to doubt everything; both save us from thinking.
—Korzybski
Wow
How interesting! I will have to order this book as well!
On point, though, and my wife would beat this into me - the portait of the girl in the tutorial - way over emphisizes her skin flaws. No woman would want a picture of herself looking like that
Dan is much more careful about women's skin flaws than I am. He does some or all of the following:
Use Filters: Blur: Surface blur to blur skin but not eyes, hair, etc. You have to play with the settings, but start with a very high threshold (everything is blurred) and find a radius which does what you wan tot the skin, probably between 2 and maybe 7. Then reduce the threshold until the detail is restored in eyes, hair, clothes, etc. I like to do this on a duplicate layer so I can further control it by using the blend-if sliders to limit the blur to the yellows and lighter parts of the images and reduce opacity as surface blur can make things look a little fake. Do this blurring early, perhaps even before the green channel luminosity blend, but certainly before the color enhancement moves (overlay blending in LAB.)
Don't sharpen the flesh or at least don't sharpen as much as other parts of the image. Dan sharpens through an inverted luminosity selection. I don't have that finger habit, actually. Do the conventional USM on a separate layer and then use the blend-if sliders again exclude the most yellow parts of the image (B channel sliders) from the blend.
I have been using the method taught in Dan's book on portraits since I first read it and to make it even faster, I created an action for it.
If anyone is interested in trying the action and letting me know how it works out for you, just send me a quick pm with an email address and I'll send it to you.
Cheers,
Mike V
P.S. If there is a way to post the action here for people to download I can do that too, just tell me how.
I am looking for these LAB Actions: Dan Margulis Portrait
DM and Swap AB. I am looking for them for Mac. I have been using this technique for the last three months with a lot of success and sometimes frustration. I of course am using a Mac and it would be helpful to run these actions and compare the results with my own. My contention with my own work is that once I get the A/B overlay done I usually end up dialing the opacity to sometimes 13% at the most. Having the actions for Mac would be valuable to me as a resource.
I am looking for these LAB Actions: Dan Margulis Portrait
DM and Swap AB. I am looking for them for Mac. I have been using this technique for the last three months with a lot of success and sometimes frustration. I of course am using a Mac and it would be helpful to run these actions and compare the results with my own. My contention with my own work is that once I get the A/B overlay done I usually end up dialing the opacity to sometimes 13% at the most. Having the actions for Mac would be valuable to me as a resource.
recipe:
1) in RGB : New Layer ; Apply Image with channel "blue" ; mode=luminosity & opacity=50%
2) go to Lab : New Layer ; Apply Image with channel "a" ; mode=overlay & opacity=50%
3) paint on noise highlight with mode=color & blend mode IF top layer (select some interval of luminance)
4) go to RGB : stamp layers; duplicate layer ; filter high pass with mode= soft light; opacity 70%
5) in option: remove shadow on the left part of portrait
Your technique is no match againts Dan's. The hair looks navy blue and overall seems to have a blue cast.
Well, it's been a long time and this portrait workflow is still with me. In fact, it serves as the framework for nearly every color enhancement I do. I think a lot of people have moved on to LightRoom or to other ways of working, but I have yet to find something else that give me the results I like in a way I understand as well.
Anyway, I reorganized a few years ago and the link to my action set for this workflow was broken. Here is a link to a new action set for the Dan Margulis portrait technique.
There are three actions in the set:
G Luminosity Blend; goto LAB -- Makes a layer with the green channel, sets the blending mode to luminosity and change mode to LAB. After this you can tweak, for example using blend-if to keep the reds from getting too dark.
Make Overlay Layers -- Makes separate A and B overlay layers. Sets the blending modes so that the topmost layer only blends the A channel. Often with Caucasian people, it may be useful to turn down the opacity of the topmost (A) layer before flattening to keep from getting skintones too magenta.
(optional) Swap A & B -- Pulls the B layer to the top and adjusts the blending modes so it's opacity can be reduced relative to the A layer, which is sometimes useful to keep dark skinned people form getting too yellow.
Comments
That's an interesting suggestion and I can see why it would have worked better in this case. The Margulis recipe summary on page 324 does not suggest this - in fact it advises the opposite (to set the blendif sliders while in RGB). But, in the several examples he goes through he sort of proves that there really isn't a recipe for this step since he uses a different method on nearly every image. That's another one of the reasons why this isn't really a recipe. You have to fully understand every step so you know how to adapt them to a particular image.
I find the image a lot more self describing when I have a copy of the green channel with a luminosity blend mode than when I have a dupped layer that has somehow been changed (via Apply Image). The former shows me exactly what was done to it - the latter does not. It also shows me the actual green channel that I was blending (in B&W) so I can see why things came out too dark if they did. The Apply Image method doesn't show me any of that after-the-fact. Since either way will work, folks can make their own choice.
Just curious about a question: Do blendif sliders set on the luminosity layer in RGB mode convert properly to LAB mode? How would they get represented in the LAB mode UI? Would a green blendif setting get converted into pieces of an L, A and B blendif setting in LAB mode?
That helps a lot to understand the curves that are the same as overlay mode. Now the obvious question. There's no magic of overlay mode. It just happens to be equivalent to these curves and convenient to use. So, why these specific curves for portraits? Wouldn't the specific curve that works best be quite image specific, just like it is when applying contrast curves to the L channel? And, wouldn't using curves instead of overlay give you a lot more flexibility in this regard?
Also, this begs the question of when should you push in the ends on the A/B curves and when should you use an S-curve and when a combination of the two? Chapter 12 sometimes does one and sometimes the other. Is there any easy way to tell which is more appropriate for a given image?
For an RGB curve in RGB mode, I know to only push in the ends when the histogram is either empty at one end (so pushing in the end just spreads out your tones into unused tonal space) or when I'm purposely pushing some things to full black or full white (more commonly done on a mask than on an image), but sometimes done at the black-end on an image to give it a bit more pop. Should I be thinking of the A/B channels the same way and looking at their histogram? Or does LAB work differently in this regard?
I'm a bit confused because I've always wondered why pushing in the ends on an A/B curve (the basic Chapter 1 technique) doesn't clip something like it does on an RGB curve.
I haven't fully adopted Margulis' sharpening techniques yet. Not because I don't like them, but I just haven't fully gotten there yet.
Thanks for the feedback.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
In truth, Dan doesn't actually suggest using curves instead of the A/B overlay trick, so the fact that the curves that mimick that proceedure look the way they do is somewhat beside the point. The feature that those curves, the channel overlays, and the techniques from chapter one all share is that they are perfectly symmetrical shifts. This keeps things safe, in terms of avoiding gut-wrenching shifts in color where you aren't looking for them.
The LAB color curves are so powerful, and sensitive, you wouldn't want to simply whip up an "S" shape, as would be appropriate in CMYK if you see that magenta is flat. The whole second half of Chapter 12 deals in-depth with non-symmetrical moves, and provides some careful guidelines for identifying when they're appropriate, and how to avoid disaster.
As for why the ends of the A & B curves don't clip, that's because your monitor's RGB gamut has already done all the clipping. The colors out at the edges are outside every gamut imaginable, and probably some that can't even be imagined. However, just as with RGB and CMYK curves, the difference between pulling an anchor point, and establishing an "S" shape has as much to do with the way the move is distributed throughout the rest of the curve, as with the point being shifted. If the lowest value in, say, my cyan channel is 16, and I have an area that I definitely want to go white, I can either set my anchor for Input=16, Output=4, or I can set a point at 16 and pull it down to 4, leaving my anchor at 0. My highlight will reach the desired value either way, buut the rest of the tonal range will be very differently affected, depending on which method I go for, as the very different shapes off the curve suggest. The first version will be darker overall; the second version will not only be lighter, but the lightening increases as the bottom point is approached.
One thing to note, using layers over Dan's method: if both A and B layers are at 100%, of course only the top will show. Reducing this to 50% will balance both channel blends, but the intensity will be at 50% for each. That's the max, for a balanced effect. If, as I have done, you place the B layer on top at, say 60%, that leaves your A layer to come in at 40%. As to whether it's a bug or a feature of using layers, I'd weigh in on the side of a feature, since every time I've tried it Dan's way, blasting the channels on top of themselves at 80—100%, I've ended up dropping the resulting layer's opacity to around 40% or thereabouts. However, it's important to recognize that substituting layers for each channel blend is not a precise duplication of Dan's method.
I'm still on the fence about this technique, as opposed to MFM curves, as opposed to simple A/B steepening, as opposed to doing it the old fashioned way in CMYK. I've gotten my best results so far through a combination of overlays and MFM curves, but I've still gone back into RGB or CMYK for some final moves that aren't available in LAB (Luminosity contrast to the magenta curve, selective color to add a little cyan to Reds and Yellows).
Where all this is heading, clearly, is yet another extention of the age-old Photoshop rule: there's always another way to do it. Trying to think in terms of this step, or that step, is shortsighted. It's more a function of this concept or that concept, and picking and choosing as conditions, and knowledge, dictate. As I've become more and more comfortable visualizing what happens when the A and B curves are transformed, I find that I am starting to see all the techniques in their proper perspective: more ways to do things to color.
—Korzybski
Let's forget about what Dan says to do. You are right, it's not really a recipe. In some sense, none of what Dan teaches is a recipe; rather it's a way of thinking about this stuff, a general approach, and a bunch of tools to implement it.
But there is more than one reason to flatten in LAB instead of RGB. LAB handles extremely light and extremely dark colors better than RGB or CMYK can because of it ability to express them as impossible colors. It's important to understand this if you are really going to understand the differences between working in LAB and the other color spaces. At this point you can just use dgrin search for impossible colors to find out more.
So, unless you have a reason not to, take the luminosity blend to LAB before flattening it.
After Apply-Image in "Normal Mode" at 100% opacity with the composite (RGB) as target and the green channel as source, the result will be EXACTLY THE SAME as if you selected the green channel and pasted it into a new layer. You can take my word for it, or you can do an experiment: Do it both ways into separate layers. Then set the blending mode to "Difference".
No, they are not converted properly. If you set them in RGB mode, you have to flatten in RGB mode.
Very good questions. I went through exactly the same reasoning. Edgework, too. He concluded that the MFM curves generally worked better than the Overlays. I've found that the overlays work really well on images with nice color balance to start with and they are pretty easy to accomplish. If there is interesting light (as with the ballet pictures) or the skin tone is too flat (not just too gray), I like the MFM method. Whatever. Really all three things are just variations on a basic simple theme: get extreme color variation in a layer and then lower it's opacity until it looks good.
Unfortunately, not that I know of. With experience and trial and error better results come more easily.
Quite differently. The L curve is like the RGB composite, except it is guaranteed not to affect color, only luminosity. Take what you know about the RGB composite curve and apply to the L curve and you'll feel pretty much at home.
Generally the stuff it clips is so far out of gamut that you can't even imagine it. Run your mouse over one of your images with the button depressed and see where the point is on the curve. You won't have anything out in that last 20% on either side.
Sharpen the L channel. The results are always better than sharpening in RGB. If you really know what you are doing, there are times when you should sharpen in CMYK. See my sharpening tutorials.
It also has a different gamma.
Dgrin FAQ | Me | Workshops
I understand and agree that the resulting image is exactly the same. I was just pointing out that it feels to me like the difference between using Image/Adjustments/Curves on a dupped layer vs. Layer/New Adjustment Layer/Curves. The latter is a lot more obvious what was done after-the-fact than the former, though they achieve the same result. We're just talking about a style choice for how one wants to accomplish the same result.
Some variations don't require extreme manipulations that you then lower the opacity. In some cases you can just adjust to taste directly. Again, just a style choice for how you want to do it.
OK, that explains it. It is clipping, but there's usually nothing there to clip. That restores my sense of understanding how curves work. I sure wish CS2 would display the histogram in the curves dialog like it does in the levels dialog, in the ACR curves dialog and like other competitive products do. That would make this a lot of this easier to see.
I'm getting there. One thing at a time. From what I've read there's not much difference between sharpening the L channel in LAB mode versus sharpening in RGB mode and then fade to luminosity. Also, is there a way to see the full color version of your image while you are sharpening the L channel? Or do you have to only look at the L channel while doing the actual sharpening?
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
Ah. Select the L channel by clicking on it. Image display goes to B&W. Click on the visibility checkbox next to the LAB composite (the eyeball). Now only the L channel is active, but you see all three channels and the image will be in color. The display in the USM dialog is still B&W, unfortunately, but you can see what you are doing with preview on the main image.
Select the L channel and then hit the tilde key. Easy-peasy.
Dgrin FAQ | Me | Workshops
Ah-ha! More hidden Photoshop secrets. Thse will make sharpening the L channel a lot better.
Thanks.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
"It is a magical time. I am reluctant to leave. Yet the shooting becomes more difficult, the path back grows black as it is without this last light. I don't do it anymore unless my husband is with me, as I am still afraid of the dark, smile.
This was truly last light, my legs were tired, my husband could no longer read and was anxious to leave, but the magic and I, we lingered........"
Ginger Jones
Anyway, here goes. I really think I went too far. In fact I KNOW I did. I ran it past Andy, and basically said WTF? So I scaled back on the saturation. I went back and pulled back on the HIRALOAM. I just pulled back overall, because I was too gung-ho. I also had a yellow cast that I wasn't careful enough with early on, and it got worse.
Before/After:
Dgrin FAQ | Me | Workshops
Before:
After:
Dgrin FAQ | Me | Workshops
Thanks. I guess I'm trying too hard to avoid *any* area in the skin where the A is more than the B. Maybe I have to relax a little and let parts of the skin be red.
And I'd like to know what you think aside from that...
How are these? (minor curves tweaks)
Dgrin FAQ | Me | Workshops
PICT2352Final2.jpg
no, you don't have to avoid A being more than B. you've done what i did with this image... missed a cast because it makes B more than A throughout the image. so i ended up with a very orange result.
even in a very pale skin, you'd at least expect the cheeks to be tending to the A.
but in my picture, i corrected the cast. that leaves his skin mainly more A than B. but checking margulis again, he says in a dark haired male with some sun exposure, you'd expect this. she's wearing makeup. and she's towards the pinkish side anyway -- a bit of a rose. and both these kids are aussies in summer and have plenty of colour so i brought the A/B blending right back. i'm often having to remove colour. i checked margulis's first example and she's very pallid. so the recipe works very well.
i can only post one pic...
Oddly enough, what is usually considered LAB's strongest feature, the ability to deal with contrast and color in totally separate operations, can sometimes be a problem with faces. In this picture the problem is not so much with the color (the numbers are certainly within range) but with the lack of contrast. Pulling the highlights in the lightness channel gives an odd result, not at all natural looking.
Good old CMYK is a powerful space to work with faces, since the three color channels are so distinct in the way that they contribute to both the color and shape of a face. In your image, both Magenta and Yellow need to be stretched downward, like so:
Then, just to make sure the overall image didn't go too cold, I added a second curve in Color mode to put back a touch of warmth. A magenta anchor at 64 moved to 66, and a yellow anchor at 64 moved to 68. I also used a selective color layer, also in Color mode, where I set Cyan for -21 in the yellows. This had no effect at all on the skin but it helped keep her hair from pushing to the greenish side.
Trying to attack the subtle shift in the skin tones is asking a bit more from LAB than it's designed to accomplish. Similar moves in the Green and Blue channel get similar results in RGB, but I think CMYK is the space of choice for fine tuning skin tones. As for the fear that color might be lost in the translation, those tones out of gamut are usually the kinds that don't look natural anyway: bright blues, greens and fire engine reds.
—Korzybski
i kinda don't know what i'm doing here...
My experience with the AB blend is that it obliterates the cyan information, which is exactly what you would expect. The A/B channels are pure color information and, in a face, they will emphasize magenta and yellow. The overlay operation simply makes whatever is there that much more so. Since the lightness channel stays the same, the overall value cannot get darker to account for the added magenta and yellow, so guess what suffers? Cyan.
I've still found it to be a really useful step—in fact, the whole recipe is useful—but in context. When I posted a sample earlier in this thread, I followed Rutt's suggetion of placing each channel's overlay operation in a separate image on a separate layer. This restricts the maximum percentage for both channels to 50%, though I set the B layer on top at 60%, giving it the edge over A. I also put both layers in a layer set that I reduced in opacity as well. A full intensity channel overlay blend in both A and B is something that I have not found to be workable, and there are other ways to achieve the same effects, with more options and subtlety in the moves. Keep in mind that these recipes are things that Dan Margulis evolved for people who want to derive some benefit from the power of LAB but who aren't really interested in getting deep into the dynamics of what is happening. A one-size-fits-all approach will necessarily compromise across the board in exchange for delivering a more-or-less decent result.
I've also found it useful, after all the LAB work is finished to take the image back into RGB or CMYK and use Selective Color on the Reds to add a touch of cyan, just to cool things off. One way or another, you will need to compensate for the drop in cyan information, however.
—Korzybski
How interesting! I will have to order this book as well!
On point, though, and my wife would beat this into me - the portait of the girl in the tutorial - way over emphisizes her skin flaws. No woman would want a picture of herself looking like that
But, this is a very interesting read!
www.shadowlakes.com
There is much much more you can do for women's skin. See: http://dgrin.smugmug.com/gallery/1170442
Hi,
I have been using the method taught in Dan's book on portraits since I first read it and to make it even faster, I created an action for it.
If anyone is interested in trying the action and letting me know how it works out for you, just send me a quick pm with an email address and I'll send it to you.
Cheers,
Mike V
P.S. If there is a way to post the action here for people to download I can do that too, just tell me how.
Hi,
I am looking for these LAB Actions: Dan Margulis Portrait
DM and Swap AB. I am looking for them for Mac. I have been using this technique for the last three months with a lot of success and sometimes frustration. I of course am using a Mac and it would be helpful to run these actions and compare the results with my own. My contention with my own work is that once I get the A/B overlay done I usually end up dialing the opacity to sometimes 13% at the most. Having the actions for Mac would be valuable to me as a resource.
Thanks!
Your technique is no match againts Dan's. The hair looks navy blue and overall seems to have a blue cast.
I'd love to contribute to the discussion, but alas I can only imitate.
Here is my favourite tutorial on LAB skin enhancer which is a great way in for us beginners:
New Portfolio
That site no longer works and refers back to a search site.
Nicely done!
Dgrin FAQ | Me | Workshops
Anyway, I reorganized a few years ago and the link to my action set for this workflow was broken. Here is a link to a new action set for the Dan Margulis portrait technique.
There are three actions in the set:
Thanks for the link also.
Moderator of the Technique Forum and Finishing School on Dgrin