Ghost in the Shell

[BronzeTitan]

I tried the point-to-point color correction approach by starting with the original artwork, . .

. . and overlaying the HD-TV (Judgment) screencap, . .

. . and selecting a variety of color points to sample (shown here only with the background pictures were dimmed, for easier visualization) . .

The sample R-G-B values from both pictures were then entered into a paint program's Curves function . .

When I saw these graphs (and the resulting HD-TV adjustment) I knew there was a problem. When near-by colors jump around like this, it's always a clue that the newer source was digitally manipulated (an HSL Wheel/Scope/whatever) rather than analogically adjusted. Segments of the spectrum were independently changed in different directions.

So the Curves technique, by itself, failed do the job of color correction. Just to continue the test, I manually pulled the those points, outside of the general curve swing, closer inwards. That removed psychedelic  "color patches" in the picture but colors were still notably off (too strong, wrong hues, and such).

Continuing with the test, I added the HSL Wheel adjustment to it. That easily brought errant colors back into line -- not perfectly, but closer to a representative of the original artwork.

So scrapping the point-to-point Curves as the main approach, I'll use the HSL Wheel. Stay tuned. 

[CSchmidlapp]

A project to return GITS to it's original cell colour's may make you promoted to a much denser metal that bronze.
And I mean able to crush Wolverine's skeleton just with the faintest of glances.

That original cell background is gorgeous, and makes me a little sad that there may never be a scan and release possible, to capture a higher resolution than we already have. Although I have seen magic around these parts.

Can not wait to follow your progress good sir.

[BronzeTitan]

Too bad we can't get in contact with "anonymous" for more originals.

Yet, even this one is not perfect. It could be the .JPG compression. More likely it was too much lighting for the photography. Either way, it has insignificant blow-out, but sizable crush in the dark alley. (R or G or B values of "0" are shown as full brightness ("255") primaries; R/G/B combinations show as full brightness secondaries; white indicates all 3 primaries were crushed) . .

That won't affect the color-correction much, as this HD-TV (Judgment) version is generally lighter and still has it's detail in those areas.

[BronzeTitan]

As a hot-shot space-pilot once said, "Sometimes I amaze even myself."

True, I expected this correction to look, you know, "correct". But I wasn't ready for Ghost In The Shell's transformation from nice cartoon to gritty realism. In a word ...  AWESOME !

One caveat here is that, to produce a really good looking picture, I gently denoised & sharpened HDTV-Judgment as demonstrated earlier in this thread. This is a legitimate procedure to prevent color-correction from emphasizing the noise, too. So, let's go for the gold ! 

Use the original artwork as a background, . .

. . and resize/align HD-TV (Judgment) . .

. . as a layer . .

STEP 1

Notice the brightness differences -- the darks are too light and the lights aren't light enough. This is a job best suited to Histogram because it keeps a smooth, general adjustment of HDTV-J when matching the overall brightness levels to the artwork.

The stone foundation needs darkening, but the only way to do it without crushing even darker areas is to increase the contrast in Midtone: expand.

The sky is another matter. It's Red is notably higher than both the Green and Blue (from using an eyedropper to read those values). As Red would be decreased later on, it should be ignored at this stage. A brightness match should be relative only between the Greens and Blues. Therefore High is reduced to pull the spectrum up. The sky, then, still looks brighter because Red is higher, but the critical Green and Blue are matched to the artwork, which is what we want.

The greys of the shadowed buildings are lowered by reducing Gamma.

Do a little back and forth to balance anything affected by the other adjustments, and then that part is done. Values to achieve this (apart from the defaults) are highlighted in yellow . .

STEP 2

Once properly prepped, the main adjustment is with the HSL Wheel,  to move all the colors back to where they belong as per the artwork's spectrum. Adjustment for each primary and secondary color-wheel will follow the same procedure.

Temporarily max the Saturation to see where HDTV-J  is affected. Then, starting at the default, vary the Hue, Saturation, and Lightness to see where that color area goes, to roughly match that of the artwork. Move on to the next color-wheel. When the 1st pass is finished, repeat the procedure in case colors are not smoothing into their neighbors. Compromise where necessary but bring everything as close to "best look" as possible. Then that's done. Again, values to achieve this (apart from the defaults) are highlighted in yellow . .

STEP 3

At this stage, you will notice some areas that look slightly different -- shadows a shade off in certain areas; color objectss still a little too tinted red, green, or blue. That's where Curves come in to make those touch-ups.

Closely compare HDTV-J  to the artwork, check them with an eyedropper, put a point to the appropriate value on the graph, and gently adjust it's change. That may also require restoring the the curve elsewhere with "anchor" points further up and/or down the graph, for those areas to remain unaltered. The RGB Curve affects the general brightness, and the individual R-G-B Curves affects the hues.

As can be seen, the graphs' lines are just a little pushed and nudged here and there. Nothing drastic. Then that's done. Values to achieve this (apart from the defaults) are imprinted, with additional ▲/▼/ ● characters to denote increases/decreases/default-anchors . .

If the corrected layer is toggled off and on, it should match really close. More could be done with tighter ranges on more HSL Wheels but this is where the paint program stops.

Here is the "before", as on Judgment's website, earlier mentioned in this thread (open the picture in a new tab to see it full-sized) . .

. . and here is the "after" to match the original artwork (open the picture in a new tab to see it full-sized) . .

[spoRv]

That's very good! Is there a way to apply the setting to the whole movie? (if, of course, a single setting would fit well)

[Doctor M]

I have to ask. Is there any reason to believe the 'original art' is accurate or as intended?
It seems a lot like Disney using original cells for color timing new transfers.

[CSchmidlapp]

I have to ask.  Is there any reason to believe the 'original art' is accurate or as intended?
It seems a lot like Disney using original cells for color timing new transfers.

The process to create GITS would have added alot of variables.

As far as I can muster the shots we're digitally created via cell scans and then a 'film out' was done for each shot for editing.
Which was then edited on a AVID, and a negative conformed to that. Consisting of the film outs mentioned!
Shockingly no Digital Master or Digital Intermediate was created! The master was on film.
I assume this master is a negative and the Inter positive to release prints journey followed.

I'm also under the impression the video master created at this time, is the same one being used to this day!
It holds limited Resolution, Colour information and has problems inherent with a video master of this time period.
It certainly has a magenta favored balance in which I'm unsure if this was just to accommodate the CRT technology or they corrected it based on the inter positive.

The 35mm print I saw had all the traits of (what I think are) characteristics introduced by a release print (Mainly Contrast and a push towards the teal). This means the interpositive's / internegative's would probably have been prepared for that, with contrast flattened somewhat and a push toward the reds in the colors. Any video masters corrected to this are going to be wrong in the sense of color timing. i think even the bulb of the projector has to be taken into consideration!

Im making huge assumptions here and massive amounts of speculation based on whats available, but I'm sure the original cells are what they we're aiming to get on screen.
The deficiency of the delivery medium's should be taken into account, and Im well up for a reconstruction of the original materials look

[Doctor M]

True.

Disney did Technicolor tests on their art to achieve the colors they intended. I get the idea that the people working on GitS let the technology lead them and really didn't know what they could do with it until they were done.

The original cell/background art probably IS as accurate as any other color choices.

[BronzeTitan]

With us having lived through George Lucas syndrome ™, it's clear that these are tough calls to make. Unless we have more disinterested parties who worked on the films, telling us they did this such for that reason, my tendency is to go as close as possible to the original sources, if they can be had, and in the best condition they can be obtained. Yeah, allot of guess-work is involved, but common sense, too. Thanks, GL, for stirring that pot.    

.

Is there a way to apply the setting to the whole movie?

How did you guess that would be my next step?     Yes, with these values, I'm going back to Judgment's website for the other screenshots he displayed there. I hope to be amazed with how well they will all blend in. If not, I've got a "plan b"  bubbling in this bronze bod (of which you'll know if you saw Jason And The Argonauts). 

Meanwhile, I've kept Tinkerbell waiting for her makeover at the Peter Pan thread. But will be back soon.

[BronzeTitan]

Here's the assemblage of all the sample frames displayed at Judgment's website, as it switches between without (lighter) and with the artwork color matching . .

It looks like a single setting could handle most of the movie. Some very dark scenes could be boosted slightly, but then the noise would increase because of no detail therein (a source issue?). Ah, if they would only go back and reconstruct it ... as was done for Pixar's Knick Knack ... minus the mastectomies, of course.