Practising with portraits
El Kiwi
Registered Users Posts: 154 Major grins
Hi all,
I recently picked up a copy of both the LAB book and Dan's latest Professional Photoshop. I've been practising with the Chapter 16 recipe particularly, and I'd appreciate some feedback on how it's working out for me.
This is what I started with:
And this is what I ended up with:
To my eye, there are quite a few things to like about the new image, and a few I don't like so much. I like the increased contrast, and the increased detail in the clothes and hair. The image is generally a lot more striking.
The things I don't like so much are things I've struggled with with this recipe. There's a fair amount more noise in the background, although in this case I don't mind too much since, well, it's the background - I don't think it detracts. There's also a lot more unevenness in the skin on the face, which has been the main problem I've experienced with this technique. Given that this image is pretty crisp to start, and my friend has really nice skin, what hope is there for the rest of my victims? This doesn't seem to affect the hand, curiously.
Here is more or less how I went about it (unfortunately I stupidly didn't note all these values):
1. First I corrected a yellow cast with the white balance in Lightroom. This shot was taken in a hotel elevator with pretty yellow lighting, after the WB correction there's no obvious cast I can see but the colour is pretty washed out.
2. I used the green layer for the luminosity blend, and then used blending options to gradually exclude the lips, since they came up pretty dark.
3. For the colour steps I used the overlay blending technique, A at 80%, B at 100%. I ended up using a pretty low opacity, I think around 35%. I find it's extremely easy to overdo this step, and I never seem to use anywhere near the percentages that Dan and others on this board use.
4. I sharpened the L pretty lightly, about 100, 0.8, 8, using blending options on the a to exclude the skin (the inverse luminosity mask didn't work well because it left out the areas of her hair with highlights). For the Hiraloam I used about 45, 10, 6. These values are for the smaller version shown here.
The main problems I've had with this technique have been:
1. It's extremely sensitive to low-quality originals. I find noise is often amplified terribly at various stages, notably the luminosity blend.
2. It's very easy to make the skin look bad even with a reasonable quality original like this one.
3. It's easy to get radioactive colours, although I'm getting the hang of this bit by bit.
Any comments and suggestions gratefully received!
I recently picked up a copy of both the LAB book and Dan's latest Professional Photoshop. I've been practising with the Chapter 16 recipe particularly, and I'd appreciate some feedback on how it's working out for me.
This is what I started with:
And this is what I ended up with:
To my eye, there are quite a few things to like about the new image, and a few I don't like so much. I like the increased contrast, and the increased detail in the clothes and hair. The image is generally a lot more striking.
The things I don't like so much are things I've struggled with with this recipe. There's a fair amount more noise in the background, although in this case I don't mind too much since, well, it's the background - I don't think it detracts. There's also a lot more unevenness in the skin on the face, which has been the main problem I've experienced with this technique. Given that this image is pretty crisp to start, and my friend has really nice skin, what hope is there for the rest of my victims? This doesn't seem to affect the hand, curiously.
Here is more or less how I went about it (unfortunately I stupidly didn't note all these values):
1. First I corrected a yellow cast with the white balance in Lightroom. This shot was taken in a hotel elevator with pretty yellow lighting, after the WB correction there's no obvious cast I can see but the colour is pretty washed out.
2. I used the green layer for the luminosity blend, and then used blending options to gradually exclude the lips, since they came up pretty dark.
3. For the colour steps I used the overlay blending technique, A at 80%, B at 100%. I ended up using a pretty low opacity, I think around 35%. I find it's extremely easy to overdo this step, and I never seem to use anywhere near the percentages that Dan and others on this board use.
4. I sharpened the L pretty lightly, about 100, 0.8, 8, using blending options on the a to exclude the skin (the inverse luminosity mask didn't work well because it left out the areas of her hair with highlights). For the Hiraloam I used about 45, 10, 6. These values are for the smaller version shown here.
The main problems I've had with this technique have been:
1. It's extremely sensitive to low-quality originals. I find noise is often amplified terribly at various stages, notably the luminosity blend.
2. It's very easy to make the skin look bad even with a reasonable quality original like this one.
3. It's easy to get radioactive colours, although I'm getting the hang of this bit by bit.
Any comments and suggestions gratefully received!
Constructive criticism always welcome!
"Everything has its beauty, but not everyone sees it." - Confucius
"Everything has its beauty, but not everyone sees it." - Confucius
0
Comments
You have to learn to trust your numbers. Had you done so, your analysis of the original would have prompted a couple of different steps.
In the light regions under her left eye (on the right side of the image) I'm reading a fair balance of magenta and yellow (35 and 39, respectively) but virtually no cyan.
Dan's portrait recipe, of necessity, enhances reds and yellows, by virtue of the fact that it darkens those parts of the a and b channel. So you get no relief for your impoverished cyan values.
Back to your original, in her forehead I'm reading 8 cyan and 51 Magenta, not bad at all. But yellow is coming in at 63, clearly overpowering the mix. Simply adjusting the white balance is going to affect yellow in the region where it's not as much of a problem, and ignore it precisely where it is a problem.
In addition, your scarf is showing symptoms of plugging up in the original, and the contrast moves you applied have exacerbated the problem. The scarf detail is almost gone, but there is an overall heaviness to the image that wasn't in the original.
There are some prep moves you should consider making, before trying LAB tricks.
The original is a bit flat, and certainly needed a contrast enhancement, but I think you could come down in the highlights a little more, and a couple of simple curves to the Red and Blue channels could help the cyan in the lower range and the excess yellow in the upper. Then, in LAB, I'd run Shadow/Highlights, using a fake black layer mask (check out the discussion in the Chapter 5 summary of Dan's Professional Photoshop book). That will give you this as a starting point to begin working on Dan's moves.
When we were doing the discussion group on Dan's LAB book here, Rutt came up with the technique of placing two copies of the image in separate layers and running the a and b overlay moves in each layer respectively and 100%: an "a" layer, and a "b" layer. In addition to giving you real-time flexibility in tuning the relationship between layers with varying opacity settings, it also serves the useful function of limiting the actual intensity of either layer to 50% max. As you point out, it's easy to go overboard here, and less is definitely more.
—Korzybski
I'll read up about Rutt's dual-layer technique again, too. One thing I didn't understand with it was the interaction between the two layers. Using his action, I end up with two layers, the a and b blend layers. The blending mode on them is set to normal - since the a layer is on top but still has a b channel, why doesn't this completely obscure the b of the underlying layer? From what I can tell, the action doesn't affect the b of the top layer, so doesn't this effectively give an unmodified b channel?
EDIT: sorry, I just answered my own question by looking at the blending options....
Thanks again for the patient advice!
"Everything has its beauty, but not everyone sees it." - Confucius
Would you mind describing the curves you used in RGB, edgework? It sounds like you only touched the R and B, so I'm guessing you boosted the B in the lower range (for the cyan) and reduced the red in the upper (for the highlights and yellow) - don't you need to touch the green for either of these?
Also - this is something I've been wondering about for a while - what is the relationship between the white balance tools in something like Lightroom and LAB curves? Is it basically the same thing? LR has the light temperature slider, and also a green/magenta tint slider, so it's obviously basically doing something similar. Is there any point in correcting white balance in these programs or is it better to just use LAB once you get the hang of it?
"Everything has its beauty, but not everyone sees it." - Confucius
Exactly. I seem not to have saved the psd file, but those are the moves I made. There's nothing precise about it, as this is just a preparatory move to bring the image into a decent range. Final precision would take place at the end. Keep in mind, LAB is not the best place to fine tune skin tone. In fact, it's awful, for several reasons. The kinds of tweaks that skin tones require are on the order of "Pull a little yellow out of the highlights, and a little magenta out of the shadows..." That sort of thing. LAB is too powerful to get that tiny.
The reason I use CMYK values, even when I'm working in LAB or RGB, is that, for skin, they're uncontaminated by shadow information. The only areas in skin that are important are usually not going to be found in the shadow range, but in RGB, your value readout is going to be a combination of both, one affecting the other. When I look at CMYK numbers, it's a fairly clean representation of what's going on with the color.
In Normal mode, if you want a balanced relationship between each layer, the top will have to be at 50%, the bottom at 100%, which would make both effectively 50%. You might go with a 60-40 split, or a 70-30, but you'll never get 80-100 or 70-70, which is possible when you simply blend plates as Dan suggests. As you discovered, it's easy to make your skin tones look like they took a plutonium bath that way.
As Rutt points out, blending modes can get around this, but due to the various algorithms Photoshop used with those modes (screen/multply with overlay, darken/lighten with soft light, etc) you are moving away from the results that Dan is shooting for. Nothing wrong with blending modes, I love 'em myself. But it is a second level add-on to the basic technique, not just another way to accomplish the same thing.
—Korzybski
Okay, I went back and reread this and it sounds like you're keeping the mode set to Normal, is that correct? How can you get 100% of each channel that way? Or are we talking about different things?
—Korzybski
Right, this is what I went back and checked, and it's what the latest copy of your actions does - the top layer (the a) only blends the a channel, I guess it just passes the others through.
"Everything has its beauty, but not everyone sees it." - Confucius
Ok, thanks for the tip. Time to start reading Professional Photoshop to learn more about CMYK, it seems. I'm just going to let all the information wash over me and hope some of it sticks
"Everything has its beauty, but not everyone sees it." - Confucius
which I like a lot more. I went for a much lighter touch with the LAB moves, too. It's amazing how much difference getting the colours right makes to the evenness of the skin tone, not just the colour. Thanks again for all the help!
"Everything has its beauty, but not everyone sees it." - Confucius
In LAB, I also used S/H to bring up the dress and to temper the highlights on her face and hair. Then DAN's AB overlay portrait trick.
I sharpened the Black channel, and then did Hiraloam on the Red channel back in sRGB.
Duffy
Duffy
I wanted to see what the changes I was making to the curves were doing, so I placed some samplers in spots that seemed to be representative of either colours that I wanted to preserve or the problems I wanted to fix. I placed them here:
The curves I ended up using were pretty much what edgework recommended:
and here we can see the resulting changes in the sample points:
...which is looking good. More cyan throughout the range of her face, and less yellow where it was a problem. And here's the result:
This has worked out really nicely for the skin tones. My only worry is a slight bluish tinge in a few spots that were neutral earlier (the top right of her scarf, and a few spots in her hair, especially just a bit to the right of her eye. I'll just hope this doesn't get exaggerated later, it's not too obvious, especially if you don't have the original to compare it too. I tried various blending options with the red channel, but I couldn't get it to work without affecting the face as well.
Next step was the shadow/highlight command, which I'd never used before. I used a black channel layer mask, another new one for me. The combination of the two seems incredibly powerful. I've read a lot of people on here saying that it can be overdone easily, so I basically tried to mimic the effect that edgework got for now, while I figure out how it works. In the end I kept the tonal range as narrow as I could, and kept the radius small (3, I think), with an amount of 30.
This has brought out slightly more of the blue tinge, especially in the scarf and the background, but I still think it's ok. Now it's time for the green luminosity step. This always seems to make her lips come up too red, so I used the following blending options:
which gave me:
I then create the A and B overlay layers, with the A at 76%:
And then I blend them with the original at 40%:
And that, I think, is it. I didn't want to sharpen this much more (at this size, anyway), and I played around with the hiraloam and couldn't avoid a fairly plasticky look, so I left it. I'm happy with it, I certainly learned a lot and it's a lot better than my initial effort above. The skin tones still look natural to me and the ratios are still good by the numbers.
As always, comments welcome!
"Everything has its beauty, but not everyone sees it." - Confucius
Hi Duffy,
I agree the blend works better than the original. With your original version, somehow you seem to have introduced a lot of noise and brought out the texture of her skin that I was trying to avoid. It could just be JPEG artifacts though, it's hard to tell - it looks like it might be.
"Everything has its beauty, but not everyone sees it." - Confucius