Workflow for Theatre Lighting

[rutt]

... about post processing. Especially Dan's new book and help from others. Look at how much of a difference it made to these once-in-a-lifetime shooting opportunities with difficult light:

Before:

http://rutt.smugmug.com/photos/45965092-L.jpg

After:

http://rutt.smugmug.com/photos/46024538-L.jpg

Before:

http://rutt.smugmug.com/photos/45964635-L.jpg

After:

http://rutt.smugmug.com/photos/46130389-L.jpg

I know there are many people who could have done a better job in post than I did, but only a few years ago, I wouldn't have know what do do with these shots and the opportunity would have been wasted.

[GraphyFotoz]

Incredible stuff Rutt.
Looks like your right there!! :thumb :thumb

[Aviator327]

rutt, Excellent retouch. Got me a copy of dan's Lab color last week and still on chapter 1. Which chapter did you use for the post- processing? Thanks.

[Adelphi03]

Wow, what an incredible difference between the before and after shots. The lighting look so much more natural. Give yourself a large pat on the back Rutt, you've done a fantastic job on those. :thumb

[rutt]

rutt, Excellent retouch. Got me a copy of dan's Lab color last week and still on chapter 1. Which chapter did you use for the post- processing? Thanks.

At this point? All. The techniques from this book build on each other until they are an entire workflow with interchangable parts.

[Viking]

May I ask a question?

LAB, is it nos possible to achive the exact same result with rgb or cmyk, just using different curves and so on? Will get my LAB book this week...

[edgework]

May I ask a question?

LAB, is it nos possible to achive the exact same result with rgb or cmyk, just using different curves and so on? Will get my LAB book this week...

In many cases, no. Removing color casts and enhancing colors in an otherwise dull image happen like magic in LAB.

[rutt]

Dan's LAB book presents an entire workflow full of interchangable parts that work well together. Together with his other book, Professional Shop this book presents a consistent view of how to use photoshop for color correction, sharpening, and retouching. Really, it isn't just LAB curves, but a workflow in which LAB plays a huge role.

Lots of things you can do in LAB you can also do in RGB or CMYK. Perhaps you can even prove that anything you can do in LAB you can do in CMYK or RGB. But it's so much easier if you use LAB for the things it's good at. I remember using CMYK curves for cast removal and light/dark points now and really I'm amazed I ever put up with it.

[rutt]

http://rutt.smugmug.com/photos/46872639-L.jpg

I've had a lot of requests from different people for more details on the post work for these ballet pictures. So I did a new one, and here's a step-by-step description of the procedure. Each image got special care, but what I did with this one is very representative of the process. Once you get the hang of it, this process goes fast enough, maybe 10 minutes/image.

All the methods I used are heavily influenced by Dan Margulis's new book, Photoshop LAB Color : The Canyon Conundrum and Other Adventures in the Most Powerful Colorspace, and especially the portrait technique from chapter 16. See: http://www.dgrin.com/showthread.php?t=18203

Raw conversion

I use photoshop cs2 and it's raw converter for these images. There are two goals for raw conversion: we don't want to lose any shadow or highlight detail and we'd like to get something close to good color balance which will make our life easier later on. To avoid clipping shadows and highlights, I used a linear curve (see the "Curves" tab) and reduced the exposure until there were no visible blown areas (make sure the preview, highlight, and shadow boxes are checked.) Getting good color balance under these stage lights is tough. I found a place on her toe shoe that I was pretty sure should be white and clicked the eye dropper on it. The result:

http://rutt.smugmug.com/photos/46865627-L.jpg

Remove the blue cast with RGB curves

In spite of our best efforts in ACR, the image is pretty blue. The faces in particular measure purple, very magenta and somewhat blue. So I pulled down the quatertones of the blue curve to get more reasonable (yellower) fleshtones. Typically, this improved the entire image.

http://rutt.smugmug.com/photos/46866004-S.jpg

http://rutt.smugmug.com/photos/46867322-L.jpg

Luminosity blend for improved depth and detail

Dan often does a careful B&W conversion and then uses it as a luminosity blend layer in order to get better shape and depth. In this case (as often) the green channel looked like a pretty good B&W conversion all by itself. I copied into a fresh layer and converted the whole to LAB, and changed the blending mode to Luminosity.

http://rutt.smugmug.com/photos/46865977-L.jpg

The one problem (?) here is that the green channel is very dark for bright reds, in this case her dress and his vest get very dark (but do gain a lot of detail). Dan's normal approach to this is to use the Blending Options of the luminosity layer to exclude the most magnta and least green parts of the image from the blend.

http://rutt.smugmug.com/photos/46866031-M.jpg
http://rutt.smugmug.com/photos/46866021-M.jpg

http://rutt.smugmug.com/photos/46866118-L.jpg

In this case, I think the green channel luminosity blend helps the red costumes a lot. Nevertheless, I followed the procedure through with them excluded at first. Later I thought better of it, went back and redid without the blending options at this point. That's what's shown in the before/after at the top.

Increase color variation and contrast

There are a number of ways to do this in LAB. For these images with the bright costumes under the interesting theater lights, I chose Dan's Man from Mars technique from Chapter 12 of his LAB book. I added a cruve adjustment layer and used these curves:

http://rutt.smugmug.com/photos/46866106-S.jpghttp://rutt.smugmug.com/photos/46866091-S.jpghttp://rutt.smugmug.com/photos/46866097-S.jpg

This produces a very ugly oversaturated result, like a bad 60s psydellic poster:

http://rutt.smugmug.com/photos/46867139-L.jpg

The Man From Mars trick is to reduce the opacity of this layer until it looks natural but still adds interesting color. In this case, I also dialed back the blending options for the most magenta (A postive) parts of the image to keep them from getting too far out of gamut:

http://rutt.smugmug.com/photos/46866101-M.jpg

http://rutt.smugmug.com/photos/46867256-L.jpg

At this point, I thought there was still potential for a little better contrast. It's still a little dark. I wanted better contrast across the faces. I flattened and then used this L curve:

http://rutt.smugmug.com/photos/46866084-S.jpg

with this result:

http://rutt.smugmug.com/photos/46866926-L.jpg

Fix blown highlights with the Impossible Retouch

This trick comes from Chapte 8 of the LAB book. Dan calls it "The Impossible Retouch", not so much because it's actually impossible, as because it uses LAB's impossibly light colors and Photoshops treatment of them.

There are some blown spots on the hands and on his left cheek and ear. It not that important for online reproduction of this particular image, but it will matter for large prints. But these are very easy to fix in LAB, so I did. It's a good idea to wait to do this until after color and contrast enhancement.

Here is a close crop of the hands and face:

http://rutt.smugmug.com/photos/46876463-M.jpg

I created a layer and painted flesh tones from nearby unblown areas over the blown areas:

http://rutt.smugmug.com/photos/46876475-M.jpg

Then changed the blending options of the layer to make the blending mode "Color" and used the Blend If sliders to limit the blend to only the lightest parts of the image:

http://rutt.smugmug.com/photos/46866076-M.jpg

Result:

http://rutt.smugmug.com/photos/46876450-M.jpg

Conventional USM Sharpening

Dan teaches the value of sharpening the L channel in LAB. You can read my tutorial about sharpening here: http://www.dgrin.com/showthread.php?t=9541

The result for this image was to bring out a lot of detail in the costumes and some around her eyes:

http://rutt.smugmug.com/photos/46866111-M.jpg

http://rutt.smugmug.com/photos/46865896-L.jpg

Hige Radius Low Amount USM Sharpening

This kind of sharpening is like the high pass filter; it brings out shape instead of details. It has an advantage in that it is easier to tell what it's going to do ahead of time. It is very important to make sure you sharpen thie L channel only with this type of sharpening. Start by selecting the L channel and bring up the USM dialog, set Amount to 500 and Threshold to 0 and then play with Radius values between about 10 and maybe even as much as 100, depending on how big the main interesting shapes in the image are. In this case, this is the faces and we are not at all close to them. The idea is to find a radius which exagerates the shape, cheekbones, nose, eye sockets in faces, for example. I settled here:

http://rutt.smugmug.com/photos/46866041-M.jpg

http://rutt.smugmug.com/photos/46866313-L.jpg

After lowering the amount and raising the threshold I arrived here:

http://rutt.smugmug.com/photos/46867456-L.jpg

Shadow cleanup in CMYK

Now the image looks pretty good, but I'd like to clean up those deep blue shadows a bit. So I took the image to CMYK (through RGB, very important) and applied this K curve:

http://rutt.smugmug.com/photos/46867277-S.jpg

Result:

http://rutt.smugmug.com/photos/46870535-L.jpg

Postscript

As I mentioned abouve, I decided to go back and redo without the blending options for the luminosity blend early in the procedure above. I like this version a lot better, but the steps were essentially identical except those blending options. Here is the final version from that redo:

http://rutt.smugmug.com/photos/46872850-L.jpg

This was a very long answer to a few simple questions. I hope it wasn't too much information.

[dandill]

Getting good color balance under these stage lights is tough. I found a place on her toe shoe that I was pretty sure should be white and clicked the eye dropper on it.When I can, I use WhiBal (http://www.rawworkflow.com/products/whibal/) and it works very well.

[rutt]

When I can, I use WhiBal (http://www.rawworkflow.com/products/whibal/) and it works very well.

I'm sure that works great when it works, but I'm not sure how practical it would have been for this. For one thing, I'm pretty sure I wouldn't have been allowed to put it on the stage. But even if I had, the lighting changed constantly and radically. I went back and checked. My custom white balance for this shot was K2700. I'm since told that modern stage lights almost always are K3200, so I went back and tried that. It was a little closer. But then I tried K3200 for this one:

http://rutt.smugmug.com/photos/46883853-L.jpg
Finished image

Here is K3200:

http://rutt.smugmug.com/photos/46904469-M.jpg

Here is my custom white balance at K7200 (I think it's actually ACR, Auto.)

http://rutt.smugmug.com/photos/46904649-M.jpg

A much more tractable starting point.

This particular ballet is sort of outerspace for mixed casts due to the very dramatic lighting.

[Tom K.]

Thank you for this explanation. If anyone needs convincing that skilled post processing can work miracles........you have proved it in this thread.

[Andy]

http://rutt.smugmug.com/photos/46872639-L.jpg

I've had a lot of requests from different people for more details on the post work for these ballet pictures.

Rutt, wonder if this would make a nice advanced tute for our tutes section? If you are agreeable, let David know.

Great stuff!

[rutt]

Of course, you can use this for a dgrin tutorial. I'm not exactly sure how widely useful it will be. But if I were going to shoot a professionally staged play, musical or opera, I guess I'd start here.

I have learned a few more tricks since I wrote this:


Fighting noise. I now use a trick I learned from Edgework. After increasing color and contrast with MFM curves (or something), I add a new curves layer. I duplicate the image, convert to CMYK. Then I use the inverted K channel of the duplicate as a layer mask for the new curves layer of the original. Gaussian blur that layer mask a lot (I use 8px). With this layer mask, I can move the dark endpoint of the curve inward quite a bit, pushing that noisy dark background into deep shadow and plugging. With a little care here, you can avoid plugging other near shadows.
More noise fighting: I use that same CMYK duplicate to create a layer mask for sharpening to avoid sharpening shadow noise. This time, no inverse, but use the same blur.
Better sharpening: This is a sneak preview of a technique from Professional Photoshop, 5th edition. USM and HIRALOAM on separate layers. Use higher amounts for both than you would normally, 500 for USM and about 60 for HIRALOAM. Then use the blending options to combine intelligently. Perhaps stack the USM on top of HIRALOAM and use 50% opacity. I like to use the blend-if sliders on the HIRALOAM layers to exclude the lightest parts. This prevents obvious halos and also prevents blowing highlights.
Color balance: I've learned that modern lighting designers tend to use two different color temperatures much more than any others. Floods are mostly K3200. Just plug that number into ACR and set tint to 0. Very often, it's a bingo. Follow spots are K7600. The Arabian Dance above works great the K7600.
Color balance: Sometimes it just seems hopeless to get reasonable skin tones. No ACR conversion gets close. Neither RGB nor LAB curves do the job. OK, time for the nuclear bomb of color balancing. Move to LAB, duplicate layer. Apply Image A channel to B channel, normal, 100%. Now all skin tones have equal magenta and yellow. In fact, all colors have equal magenta and yellow (and equal green and blue, for that matter.) The flesh looks right. But things that should really be blue, green, yellow won't be. So then the challenge is to find a blending mode and/or mask to fix that.

[01af]

I've had a lot of requests from different people for more details on the post work for these ballet pictures. So I did a new one, and here's a step-by-step description of the procedure. [...] Great work! But there's one detail that in my opinion could use some improvement. In the first step after raw conversion---removing the blue colour cast---the dancers' rendition was clearly improved but the blue background became pretty muddy. And then you never cared to fix that. I feel the image would look even better if you'd bring back a bit more brilliance, or saturation, into the background.


Lab---is it not possible to achive the exact same result with RGB or CMYK, just using different curves and so on? Yes, of course. The final image is an RGB image, so you can achieve every possible result within the RGB colour model, too ... in theory. However in practice, it often would be extremely time-consuming ... or virtually impossible. It's like traveling from New York to Los Angeles. You don't really need an airplane---you can walk, too. If you're patient enough, you possibly will arrive one day.

-- Olaf

[rutt]

Great work! But there's one detail that in my opinion could use some improvement. In the first step after raw conversion---removing the blue colour cast---the dancers' rendition was clearly improved but the blue background became pretty muddy. And then you never cared to fix that. I feel the image would look even better if you'd bring back a bit more brilliance, or saturation, into the background.


It would be very easy to keep the blue in that background. It's so much darker than the faces that the blend-if sliders could easily limit the blend. The inverted red channel could alternatively serve as the basis of a layer mask.

Personally, I was looking for a more neutral background. That final trip to CMYK was aimed at neutralizing it further.

[jarnold439]

John,



I am attempting to follow your workflow example that you describe on this thread. Here is what I am doing. I first use a curves adjustment layer to correct for color casts and expand the range in the interest area. Then I am merging the visible layers into a new layer (You don’t specifically say to do that but I am not sure how else to do it.) I then apply the green channel to the merged visible layer. I then change that layer to luminosity blend mode. When I go to change to LAB it discards my curves layer and messes everything up. So I figure that I am doing something wrong. Can you see where I am getting confused? You must be doing something different.

Thanks,

John Arnold

[rutt]

This is not different from the steps in Dan's portrait technique (http://www.dgrin.com/showthread.php?t=22524). In RGB, make a duplicate layer. Apply the green channel to it. Change blending mode to luminosity. Convert to LAB. Refuse offer to flatten.



John,



I am attempting to follow your workflow example that you describe on this thread. Here is what I am doing. I first use a curves adjustment layer to correct for color casts and expand the range in the interest area. Then I am merging the visible layers into a new layer (You don’t specifically say to do that but I am not sure how else to do it.) I then apply the green channel to the merged visible layer. I then change that layer to luminosity blend mode. When I go to change to LAB it discards my curves layer and messes everything up. So I figure that I am doing something wrong. Can you see where I am getting confused? You must be doing something different.

Thanks,

John Arnold

[jarnold439]

So if you use curves early on to control casts, you have to flatten before the luminosity blend. Correct? Otherwise when you convert to LAB you lose the curves adjustment layer.

[rutt]

So if you use curves early on to control casts, you have to flatten before the luminosity blend. Correct? Otherwise when you convert to LAB you lose the curves adjustment layer.

Yep.

[momwac]

I'm a little confused here. The enhancements in sharpness and detail are very impressive :clap and I hope to learn a few things by studying the detailed example. But as to color casts -- why would you override the lighting director's intent? :dunno

Edit: Aw nuts, how did I manage to resurrect a two-year-old thread? Sorry...

[rutt]

I'm a little confused here. The enhancements in sharpness and detail are very impressive :clap and I hope to learn a few things by studying the detailed example. But as to color casts -- why would you override the lighting director's intent? :dunno



Because we don't see the scene the same way we see the picture of it. If I left the cast the picture would look very blue and wrong after I enhanced the color later in the workflow. And our memory is definitely different from what the camera captured.

Here is a little theory of mine. Our visual system is powerful enough at neutralizing colored light to allow us to gauge whether a person has healthy skin tone. That means that at least some of people's flesh should be more yellow than magenta and neither green nor blue. When I want to show a cast, I let some flesh vary from the rule of thumb, but if it all varies, than the shot doesn't look right.

http://rutt.smugmug.com/photos/69321630_Fv2ze-XL.jpg

This is from Balanchine's Serenade at Boston Ballet. This ballet traditionally uses blue light and I wanted to show it. But if I don't adjust the cast enough to have some healthy flesh tones, it looks like a cast instead of colored light.

Here is a different example.

http://colortheory.smugmug.com/photos/153552178_zNF9t-L.jpg

Kathleen Breen Combes, Giselle, Boston Ballet


This is straight out of the camera jpeg with auto white balance (I do have the raw). Let me tell you a few things about this scene before I continue. The subject is dancing the part of The Queen of the Wilis. She is a sort of ghost, the spirit of a dead girl betrayed in life by a man. The scene is moonlit. To convey this, the traditional lighting is very very blue. The dancer wears heavy makeup to convey a deathly pallor. And the dress is actually blue, not white. All this to overcome our visual systems' natural tendency to see healthy flesh tones in such an obviously fit, strong, and young person.

Fine. But does the scene actually look this blue to us? Not to me anyway. I struggled for years about what to do with this shot. I want to show the light, but I want believability. Mikko Nessinen, the artistic director of Boston Ballet suggest B&W, but I thought that wouldn't address the most interesting issue: how to show the light.

Eventually, I did this:

http://colortheory.smugmug.com/photos/153965064_Jzc6R-L.jpg

Instead of healthy flesh (she is supposed to be dead) or neutralizing the white dress (it's actually blue) I settled for some color variation so it doesn't look like a duotone. I also used a layer mask to get neutral blacks. I think this conveys the scene as I remember it pretty well.

An interesting side note. I posted this as a case study on Dan Margulis' Applied Color Theory mailing list. A lot of very good prepress people took a swing at it and got very different results. Of all them, I like mine best, but not because it's mine. Seriously, take a look here (http://colortheory.smugmug.com/gallery/2856842_CQMtt#153552178_zNF9t). It's possible to make this scene look like a healthy person in bright sunlight. Once we learn how to do that, the issue comes down to what we want it to look like, not how we get there.

[321Shooter]

Hi Rutt,

Can you tell me where this book is available? I just kinda fell into this discussion and have no clue what book is being mentioned. Thanks!

[rutt]

Look for Dan Margulis on Amazon. I recommend staring with the LAB book.