Annie Leibovitz and Norman Jean Roy looks

[Sara86]

I know this has been doing the rounds recently, but the Leiblitz preset from Applied Image is one of those things I've taken to using on everything, even if only at 2%.

I'm going through my pictures trying to find something suitable to show it off, but nicked this from the Model Mayhem forum instead as although a published image it demonstrates it more clearly. (before then after)

http://i.imgur.com/jBXmHdV.jpg

I only use the top preset, as it does everything I need, then add my own processing. The effect is clear on Julia Roberts' hair. I think of it as instant color contrast. Enhances colors by offsetting them subtly against their complimentary colors, and you can easily take from warm to cool tones. Just a heads up and interested to hear how others use it.

http://appliedimage.co/

[skoobey]

Re: Annie Leibovitz and Norman Jean Roy looks

You need a preset to lift a black point on a blue channel curve up a couple of points?

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

If it can do it better than I can do it, I'll use whatever I can find

Maybe too subtle, raiding Model Mayhem for better examples.

http://i.imgur.com/gjgjr1n.jpg

http://i.imgur.com/rZBkwCy.jpg

After and Before:

http://i.imgur.com/edkeHUo.jpg

To me it's slightly magic, and saves a lot of masking and work with curves. Probably one for people who get asked to do this look a lot and have trouble getting it right. I understand this look is actually always done with a custom ICC preset.

[klev]

Re: Annie Leibovitz and Norman Jean Roy looks

To me it's slightly magic, and saves a lot of masking and work with curves. Probably one for people who get asked to do this look a lot and have trouble getting it right. I understand this look is actually always done with a custom ICC preset.

It's possible, but you would have to really stress test that thing prior to using it. You could apply a tweaked ICC profile as an input profile during raw processing, although I don't think the shops that work with Liebovitz or any of Roy's retouchers (he had a small group in house when he was in LA, not sure now) would do this. In both retouching still or motion imagery, practitioners try to avoid things that can't be undone with minimal work.

You can create custom input profiles for raw processing, which might be the way to go. The other method would be an even bigger hack. You would output to the constraints of a particular color space, then take those values as coordinate values within a new possibly unrelated color space.

The first differs in that you're actually biasing the model. The camera doesn't have an inherent color space until you define a model for it, which approximates the behavior of the hardware to the rest of the system in a comprehensible manner.

[skoobey]

Re: Annie Leibovitz and Norman Jean Roy looks

Why even go into that kind of overthinking? It's brightness contrast, then a yellows selective color that raises the saturation of the yellows and lowers the cyan, topped of with a play with red and blue black points on a curve. There is 0 need for a preset, nor does it do anything that you can't make an action for on your own in 5 minutes.

I just don't like when simple things like this are being presented as a find of the century.

EDIT: I am not bashing on what those guys did, their work is good, website looks good, it's just that for a retouch pro... we can do better than struggle with the simplest of presets. I see this floating around on MM and DP, and that is a real place for this sort of trickery IMHO.

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

I've got to confess the studio I was formerly with went to some lengths to recreate the Roy look for a few clients. We had a before and after or two, and even managed to obtain one of his PSD files, which had some very unusual Curves.

The missing bit was an oddly named ICC profile that may have been based on a film process, or may have been a way of processing drum-scanned film, just going on how extreme the Curves were. We ultimately had to do quite a bit of masking to get the look without just getting a garish crossprocessed effect, because as you say there is a lot of cyan pushed into the shadows and yellow into the highlights, but it's working quite differently on skin tones and stronger colors. I suppose it avoids the mud you'd normally get with things like strong reds.

I'd never even thought of creating custom input profiles before. I'm not sure what software allows you to do that? I guess it would be like unbounded LUTs, but obviously camera specific? This preset uses a LUT for the heavy lifting, then some fairly simple color shifts. At my previous place we had Actions presets for everything. The volume and turnover got so high. It's partly why I went freelance.

[klev]

Re: Annie Leibovitz and Norman Jean Roy looks

I've got to confess the studio I was formerly with went to some lengths to recreate the Roy look for a few clients. We had a before and after or two, and even managed to obtain one of his PSD files, which had some very unusual Curves.

The missing bit was an oddly named ICC profile that may have been based on a film process, or may have been a way of processing drum-scanned film, just going on how extreme the Curves were. We ultimately had to do quite a bit of masking to get the look without just getting a garish crossprocessed effect, because as you say there is a lot of cyan pushed into the shadows and yellow into the highlights, but it's working quite differently on skin tones and stronger colors. I suppose it avoids the mud you'd normally get with things like strong reds.

The guy has been around for a while. It's fully possible that he stuck with film for a long time. Extreme curves are possible in that case. If he drumscanned color negative film, the output would look extremely flat. You actually have to be careful with that stuff. It's easy to block up shadows and stuff, even if the detail is there.

As for masking, I'm not that surprised. It's fairly typical for striking looks to require some amount of masking. The software looks like a set of presets though, and I think you could come up with those on your own. I kind of wonder if anyone has tried to train a neural network to understand old film color palettes. That would be neat. Right now we have cheesy stuff like this. I imagine within the next year or two, we will start to see some more subdued stuff.

The software you mentioned at the beginning isn't anything special. It appears to sell you curve presets, which is quite silly. VSCO does some of that, but they also bundle custom input profiles.

I'd never even thought of creating custom input profiles before. I'm not sure what software allows you to do that? I guess it would be like unbounded LUTs, but obviously camera specific? This preset uses a LUT for the heavy lifting, then some fairly simple color shifts. At my previous place we had Actions presets for everything. The volume and turnover got so high. It's partly why I went freelance.

Unbounded is a gibberish term, and I think it's rather confusing. It refers to the use of floating point numbers in image processing applications, where the 0 to 255 range (in integer terms) is approximately mapped to the range 0.0 to 1.0. It's referred to as unbounded, because the encoding can represent a much larger range, and image processing applications typically treat these otherwise "out of bounds" ranges in a well defined manner. This includes ranges that would produce negative numbers in the xyz space. They shouldn't be defined, but they can come up in practice.

Input profiles have no relation to this. In terms of where they come in, Lightroom and camera raw used to offer some choices based on Adobe's recommendations and the camera manufacturer's original settings. Capture One used to (and may still) offer a few choices per camera. You could also create your own, but it's more difficult than you might expect.

An input profile acts as a color model. It represents an approximation of the range of outputs achievable by the input. It is an intermediate description of the range of the input.

An LUT is actually a mapping. It doesn't actually have to describe any underlying relationship between images, devices, etc. It's just a list of inputs to outputs. I hope the difference is somewhat clear.

[Guest]

Re: Annie Leibovitz and Norman Jean Roy looks

There is a backstage video of Paolo Roversi where he is tethered shooting and images appear on the monitor with the style above mentioned. Could be that a sort of input profile (device link or abstract) is involved just to make the client happy during the session. I assume the images then go by traditional retouching and the input effect is undone.

But otherwise one can be just happy with density masks and apply image (calculations) to isolate part of the image to color correct.

[skoobey]

Re: Annie Leibovitz and Norman Jean Roy looks

I would think that custom ICC profile has to do more with where they intended to print the work in the end, rather then how to get to it on the monitor, but I could be wrong.

What do you consider to be an extreme curve, Sara? Like a roller coaster or?

But otherwise one can be just happy with density masks and apply image (calculations) to isolate part of the image to color correct.

Now, I just can never get to that point. If you want to affect the specific brightness range, it always ends up being a choppy mask, due to the clipping necessary not to affect unwanted areas. Sure, it's a gradient, but a very choppy one. Blurring or adding median noise only makes it blotchy.

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

The guy has been around for a while. It's fully possible that he stuck with film for a long time. Extreme curves are possible in that case. If he drumscanned color negative film, the output would look extremely flat. You actually have to be careful with that stuff. It's easy to block up shadows and stuff, even if the detail is there.

As for masking, I'm not that surprised. It's fairly typical for striking looks to require some amount of masking. The software looks like a set of presets though, and I think you could come up with those on your own. I kind of wonder if anyone has tried to train a neural network to understand old film color palettes. That would be neat. Right now we have cheesy stuff like this. I imagine within the next year or two, we will start to see some more subdued stuff.

The software you mentioned at the beginning isn't anything special. It appears to sell you curve presets, which is quite silly. VSCO does some of that, but they also bundle custom input profiles.



Unbounded is a gibberish term, and I think it's rather confusing. It refers to the use of floating point numbers in image processing applications, where the 0 to 255 range (in integer terms) is approximately mapped to the range 0.0 to 1.0. It's referred to as unbounded, because the encoding can represent a much larger range, and image processing applications typically treat these otherwise "out of bounds" ranges in a well defined manner. This includes ranges that would produce negative numbers in the xyz space. They shouldn't be defined, but they can come up in practice.

Input profiles have no relation to this. In terms of where they come in, Lightroom and camera raw used to offer some choices based on Adobe's recommendations and the camera manufacturer's original settings. Capture One used to (and may still) offer a few choices per camera. You could also create your own, but it's more difficult than you might expect.

An input profile acts as a color model. It represents an approximation of the range of outputs achievable by the input. It is an intermediate description of the range of the input.

An LUT is actually a mapping. It doesn't actually have to describe any underlying relationship between images, devices, etc. It's just a list of inputs to outputs. I hope the difference is somewhat clear.

It's way outside my technical scope. Someone who worked with Roy once told me he started with film, but now uses (maybe) Phase One or Hasselblad digital backs now. But the custom ICC they think stays the same.

A lot of these guys are very untechnical, so I wonder if someone's mapped a coloring process to an ICC for him and his people to work with more easily. But how it differs from a LUT in practice I have to confess I'm still not entirely clear.

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

There is a backstage video of Paolo Roversi where he is tethered shooting and images appear on the monitor with the style above mentioned. Could be that a sort of input profile (device link or abstract) is involved just to make the client happy during the session. I assume the images then go by traditional retouching and the input effect is undone.

But otherwise one can be just happy with density masks and apply image (calculations) to isolate part of the image to color correct.

There is a reverse-engineered Roversi effect or two in the pack. I've not played around with them much but they are very similar in shifting neutral color mid-tones towards cyan and desaturating the skin.

It's not inconceivable the file we had could have been from a shoot where they have the preset look either baked in on set. Whether the profile added the effect or adjusted for it? I've seen very retouched looking shots on videos with Annie Leibovitz shooting tethered. I think the Keith Richards shoot for example. With the Rembrandt-like shadows and blue and yellow tones, that looks like it's had an hour's recoloring done on it.

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

I would think that custom ICC profile has to do more with where they intended to print the work in the end, rather then how to get to it on the monitor, but I could be wrong.

What do you consider to be an extreme curve, Sara? Like a roller coaster or?



Now, I just can never get to that point. If you want to affect the specific brightness range, it always ends up being a choppy mask, due to the clipping necessary not to affect unwanted areas. Sure, it's a gradient, but a very choppy one. Blurring or adding median noise only makes it blotchy.

More like the black point at 80 and the white point at 120. All flat from recollection. Different layers for Red, Green and Blue. Like the image he was importing was as flat as RAW video, with the black and white points all over the place.

I'm fairly sure it was not a technically sound way to work. The typical before image was a lot like the second image I posted, with everything quite muddy orange, and not anywhere near as flat as this processing suggested. clearly not white balanced. The afters have this effect of lots of cool tones in the shadows and background, yellows in the highlights and a warm skin. The only thing I'm sure they were doing somewhere along the line was isolating skin tones and applying a sort of crossprocessed effect to everything else.

[klev]

Re: Annie Leibovitz and Norman Jean Roy looks

It's way outside my technical scope. Someone who worked with Roy once told me he started with film, but now uses (maybe) Phase One or Hasselblad digital backs now. But the custom ICC they think stays the same.

A lot of these guys are very untechnical, so I wonder if someone's mapped a coloring process to an ICC for him and his people to work with more easily. But how it differs from a LUT in practice I have to confess I'm still not entirely clear.

These guys could be using trial and error. They could have some very technically inclined assistants. They could be doing a lot of things. The only thing I'm sure of is that their group has some collective knowledge of what worked well without creating reproduction nightmares.

As for ICC profiles vs LUTs, they aren't fully distinct. ICC profiles can contain LUTs. They represent a specification for describing color spaces. The color spaces may be abstract (Adobe 1998), a generic hardware description (the sRGB family, implementations differ by small amounts), or something more concrete, like the characteristics of a canon 5D outdoors mid-day. These things are described with respect to a reference color space, such as XYZ or LAB. Discussing LAB is more complicated, so I omit it.

An LUT isn't primarily a model. It just defines a set of outputs for a set of inputs without any implicit context. Most RGB profiles don't use that. They use matrix transformations, so to convert rgb1 to rgb2 you have <rgb2> = (A2) * (A1)^-1 * <rgb1> where A2 maps xyz to rgb2 and A1^-1 maps rgb1 to xyz.

An LUT on the other hand just takes a set of inputs and directs them to outputs. If you have an LUT, and you give it that same <rgb1>, you will receive an output which only depends on the LUT used. It does not take the source of the data into account as long as that data is properly formatted and encoded (otherwise that's an error).

This last part seems a little unclear. I'm saying that if given a particular piece of software such as photoshop and an LUT, the output values of the LUT do not depend on the profile assigned to the image, whereas a matrix based transformation (typical for converting between color spaces in photoshop) is dependent on the origin and destination spaces.


Does that help?

[Sara86]

Re: Annie Leibovitz and Norman Jean Roy looks

I think if I look up Matrix transformations I might just get it! I'm essentially seeing one as more contextualized and the other decontextualized? And as mentioned VSCO uses ICC profiles presumably to remap linear colors to nonlinear film colors like a LUT?

We used to do quite a bit in LAB back in the early days. I remember steeping the curves on A and B channels was a preferred way to control saturation.

[klev]

Re: Annie Leibovitz and Norman Jean Roy looks

I think if I look up Matrix transformations I might just get it! I'm essentially seeing one as more contextualized and the other decontextualized? And as mentioned VSCO uses ICC profiles presumably to remap linear colors to nonlinear film colors like a LUT?

We used to do quite a bit in LAB back in the early days. I remember steeping the curves on A and B channels was a preferred way to control saturation.

RGB effectively represents an infinite number of color spaces. Matrix transformations give you a way to convert between these in a principled manner by referring to each RGB color space with respect to the same (non-RGB) reference space. To do so, you need to be able to convert to and from any RGB space and the reference space. To do this, you need a set of equations

a11*X + a12*Y + a13*Z = r
a21*X + a22*Y + a23*Z = g
a31*X + a22*Y + a33*Z = b

You can represent the "a" coefficients by a matrix A. If you have matrix A, and you have x,y,z, then A*[x,y,z]^T = [r,g,b]^T (T is for transpose here, meaning they should be in column form. If you want you to convert the other way, you need A^-1 * [r,g,b]^T. Since they're only 3x3 matrices, this isn't a big deal. Here's a popular page on this subject.

The term non-linear is somewhat contextual, and it may not always align with what you're thinking.

VSCO applies these things in Lightroom as input profiles to raw images whenever possible. It's not quite the same as applying a tweak in photoshop, as it's applied to an input state based on the hardware used to create the input and the desired result.

Still though, I suspect if we see a lot of developments in this space, it will be taking something akin to a machine learning approach.