PS LAB Color, Chapter 16 -- Recipe for portraits
This chapter presents a recipe for portraits which Dan claims drastically improves their belivability. The recipe is easy to learn but it has a few decision points which I've found difficult to master. I've been practicing it for a few months now and also practiced in Dan's class. Now I'm getting consistently improved portraits, though I still make mistakes. Dan demonstrated with about 5 or 6 images in class and competed with his students on another 4 or five images and his results were consistently fantastic. Even if you follow the steps exactly as Dan presents them, the whole recipe can be applied in a minute or so, once you get the hang of it. I've introduced some automation which replaces a few of the repetitive steps and makes it even simpler to do.
[size=+2]The basic recipe:[/size]
- Remove any cast. Dan says to do this in RGB, but really you can use your preferred method, good raw conversion, RGB curves, LAB curves, whatever. At the end of this step, your fleshtones should be warm, positive both A and B. (At least as much yellow as magenta.) The rule that B should be at least as positive as A is only a rule of thumb. Rosy cheeks, sunburn, makeup, cold wet days, sunset light, can all lead to exceptions. But in principle, it's good to find at least some flesh where this is true, perhaps on the neck. For darker complexioned people, the balance will tend to favor yellow and for lighter complexioned people it will tend toward balance with more magenta than yellow in the rosy cheeks, etc. Balancing on a known neutral area, perhaps a white shirt or grey hair often works. Important:Don't worry if you desaturate a bit at this point. As long as the fleshtones are warm, the recipe will add plenty of saturation soon.
- Improve definition, depth, and sharpness by using the green channel as a luminosity blend. Examine the each of the red, green, and blue channels indivitually. Which is the better B&W version of the portrait? Usually the green channel is best by far and much better than a naive conversion to grey scale. So just use it for luminosity.
Here's an easy way to make a luminosity blend from the green channel. Create a duplicate layer and select it. Image->Apply Image, and select the green channel of the current image. Change blending mode to luminosity.
If your image has dark reds or purples, you'll have a problem at this point because they will be black in the green channel. Use blending options to exclude them from the blend.
There's lots more to say about this step. Suffice to say, if you are good a B&W conversions, you'll get the best results by making the best B&W conversion and using it for the blend instead of the green channel.
- Convert to LAB and flatten. Note the order.
- Increase color saturation and variation Create a new layer and overlay blend it's A channel to itself and B channel to itself, varying the opacity of the blends for different complexions. This is one of the key decison points of the technique and the place I've had the most trouble. Dan rule of thumb is (a) fair skin, blue eyes, blonde hair: maybe 80% A opacity and 100% B opacity; (b) Dark skinned race: maybe 100% A opacity and 80% B opacity; (c) Else, equal (100%) opacity for the overlay blends.
Dan's way of doing this is to use Image->Apply Image for each of the A and B channels with overlay mode selected and vary the opacity. I have an alternative which I think is a little easier to use unless you are Dan and have a lot of practice with the technique. See below.
- Dial in the level of saturation and color variation introduced in the last step with the opacity slider for the new layer. Obviously, this is also one of the key decision points, but it's lots easier than the step above since you can just use your eyes to see what looks best to you.
- Attempt an L curve move on the reduced opacity layer to increase contrast and establish better white and dark points. Because the layer has reduced opacity, it isn't possible to blow the highlights or plug the shadows by doing this, so it may be an easy way to get better facial detail. Don't try too hard. Often no move is fine or the best curve is a simple 3/4 tone bump.
- Sharpen the L channel, usually both conventional sharpening, but avoiding flesh, and high radius sharpening.
- Optionally, convert through RGB to CMYK and establish a deep rich shadow without plugging by steepening the K curve.
We need to discuss the decision point at step 4.
Prroably this seems like quite a lot of work, but, as I said, I have some automation which can help and it does go very fast once you get the hang of it. First I'll introduce the automation and then I'll walk though the process with an actual image.
The automation is a Photoshop action which replaces steps 2-4. Well actually it replaces steps 2, 3, and about half of 4. You still need to make the decision of how much A overlay vs B overlay to use. You can pick up my action here. It's a set containing two actions:
- Dan Margulis Portrait
- DM Swap AB
Once you get the balance right, you just merge down the top two layers and proceede with the recipe at step 5.
Warning: My action doesn't leave room for the blending options adjustment in step 2. This is something I'd like to improve and probably either I will or someone will eventually.
Edit: Changed the link above to reflect a new more permanent location. Also the new action set contains a green luminosity => LAB action. I'll post a little more about it at the bottom of this thread.
[size=+2]Step by step example[/size]
OK, enough words, time for some action! Let's walk through the recipe with a portrait of mine, selected in part because it's easy. Here is the original:
If you actually look at the indivdual channels of this, there is a good case for the blue instead of the green channel as the best B&W version. The eyes are very drammatic, but it would bring out the freckles in a big way. A careful blend of the green and blue channels would probably yield the best result here, but for now we'll just follow the recipe blindly. In fact, I just ran my Dan Margulis Portrait Photoshop action:
and ended up with this LAB image:
At this point the layer palette looks like this:
OK at this point, some real by the numbers analysis is called for. The image is way too saturated, but is the color balance right. Your eyes won't tell you, but the Color Sampler tool will. I checked a number of places on her face and found that the there was as much yellow as magenta nearly everywhere on her face except those pink cheeks and lips. This girl just ran a cross country race, so you'd expect some pink in her face. Anyway, typical values are:
- Between the eyes: A=22, B=27
- Tip of the nose: A=21, B=29
- Neck in sunlight: A=21, B=30
- Freckle: A=46, B=52
- Rosy cheek in sunlight: A=43, B=31
- Reddest, darkest part of the cheek in shadow: A=28, B=0
So now we merge the top two layers and join the recipe at step 5. I played with the opacity slider of the top layer and arrived at this:
by using a 56% opacity on the overlay blend layer:
We are now at step 6 of the recipe, which calls for possible application of an L curve to enhance contrast. This image doesn't have a true highlight, I don't want to take that reflection on her forhead all the way to white or an impossible color; but the left eyelashes have a place that should be a true shadow. I used this L curve (still on the reduced opacity overlay layer) to open up the contrast:
And arrived here:
At this point I flattened in preparation for sharpening. This is a slight divergence from Dan's recipe in that he sharpens on the lowered opacity layer. I find this confusing, perhaps because I'm used to sharpening according to my own habbits. If you haven't already, you may want to read my basic USM tutorial here.
First I used conventional L channel USM sharpening on the L channel to bring out the gleam in her eyes, to make the eyelashes and eyebrows sharp. Since she is such a young healthy girl, I didn't have to worry about confining sharpening her skin. I'm sure this issue will come up with other examples and those who've been keeping up will be able to think of lots of ways to manage it. But for this image it was no problem. Here are the sharpening parameters:
and here is the result:
The final step is HIgh RAdius LOw AMount (HIRALOAM) sharpening (also on the L channel, very very important.) This takes a little practice, but it's a great thing to know how to do. Use the Unsharp Mask filter on the L channel and start out with the Amount slider set to 500 (all the way up.) and the Threshold slider set to 0 (all the way down.) Bring the Radius slider to 10 and keep increasing it until the large facial features (cheek bones, chin, eye sockets) are outlined. Too far and the face will start to go white. To little and you won't get the outlines. The right radius is very dependent on the image. Closer crops and higher resolution calls for higher radius. I've used values anywhere between 10 and 100 recently. For this image I settled on a radius of about 66:
At this point, don't expect the image to look good:
But you can see that is emphasizing the overall shape of here face by outlining the cheeks, chin, eyes.
Next, lower the Amount slider, until the obvious sharpening artifact become invisible. Start with 50 and try values ranging from about 40 to about 65. In this case I settled on 48:
with this result:
This will still look a little too harsh, but there are no obvious visible sharpening halos. Use the preview check box often while you do this to see how you are doing.
The final step is to soften the effect of HIRALOAM sharpening by raising the Threshold slider a bit. HIRALOAM sharpening is very sensitive to threshold, so be careful here. Too much and you'll lose the effect. Too little and it will look harsh. I settled on 6:
and here is the final state of my portrait:
Please compare it to the original and see what you think.
If not now, when?