Editing uncalibrated scan negatives with consistency

[Korbel]

I have found a way to do something like a scanner calibration by editing in photoshop. I was getting wierd colors sometimes even with IT8 calibrated scanner.

(Before, after - Mamiya RB67, Kodak Portra 800)


The process is inspired by gum dichromate color printing, where you split the image in color channels and print it one by one. But making it work was half logic, half trial and error and half intuition.

The whole process can be made with a macro, so it is just one click. Then you need to remember few numbers and you just copy the settings to every image.

You should start with 16 bit scan, any editing of colors in 8-bits is quite pointless.

As a base layer I start with 50% gray

Then I reconstruct the image from individual channels for red, green, blue.
Then I add their complementary channels for cyan, yellow and I treat magenta channel as purple (magenta is a construct of our vision, the color doesn't have wavelenght) Each channel is in solid color, blend mode normal, opacity 10%. Purple opacity is at 20% because it complements the green channel and the greens from yellow and cyan channels. Each complementary channel has inverted color and mask.

(You could add as many color channels as you like by using blend mode difference for the masking, then the parts of the image containing the selected color will turn dark.)

Then I take the original image as a key on top, blend mode luminosity. (Interestingly it has different results if you use color image or its BW version. No idea what photoshop does there)

Under the key layer I add saturation (I add a little more to bright parts)

So then you have a tree like this.

From top: Original scan - hue/sat bw clipping layer for the key - key is original image with luminosity blend - then saturation layers (curves is saturation layer too in this case) - individual color channels with masks - 50% gray

And you get image like this, which is quite different from what we started with. Thats because we have added more color channels, for instance absence of green implies presence of purple and not just gray.



Now I can target each channel individually without affecting the other channels, also I got rid of magenta (which you can bring back if you want, I have found treating magenta channel as purple and treating purple channel as magenta works best for me)

So now I have two ways to edit each color channel. I can either manipulate the channel directly, or manipulate its complementary channel and this is very powerful for any precise image editing technique.

This one is too yellow, so I just turn down the Fill of the layer down until the yellow cast is gone (fill and opacity behave the same here, but fill is now a more finess adjustment), Then the image is purple-ish, so I turn down the purple channel fill. Or I can boost the greens, it will have slightly different effect.

Then I just write down the fill % settings down, press macro and do the same for next image. In most cases simple copy paste of the settings works great and in the worst case it is really simple to adjust it.

And the best thing is you can apply the same process to digital files too, so it is quite universal. You will end up with a slightly different colors, there will be colors where none were recorded, but it generally works and my feeling is I can this way produce much more naturally looking grading.

I have also applied the macro to my old digital files and it magically fixes the color grading in most cases.

To run the macro you first need to rename the scan to "original"

Download: https://marekkorbel.cz/wp-content/uploads/2025/11/channel split.ATN

Let me know if it worked well for you too!

 

[250swb]

I took your offending image and opened it in Photoshop and pressed the 'Auto Color' button and the result came out pretty much identical to your corrected image. I admit, it's maybe not as much 'magic' that a photo forum needs, but it works.

 

[Korbel]

@250swb oh great, I wasnt even aware such a thing is there that is indeed an easier solution

 

[_T_]

There’s an adjustment layer that lets you do pretty much the same thing with a little bit of extra flexibility all in a single dialogue. It’s called “Selective Color”

 

[koraks]

Then I just write down the fill % settings down, press macro and do the same for next image.

This works only if you have scans that are consistent to begin with. How do you ensure this in your workflow? There are a few ways of doing it, but it's important to include this as a condition for your approach to truly result in consisting processing. The 'before' image suggests that you are working with a scan that's already inverted and to a large extent color balanced. The latter step virtually always uses a content-based algorithm; i.e. the result will vary with the image content, and thus, no two frames will be the same. They may be close, depending on the color balancing algorithm used, but even in the best situations you generally end up with clearly visible differences in color balance across frames. As said, there are a few ways to avoid this, so my question is which way you've used to ensure frame-to-frame consistency.

Another caveat is that this will (should) indeed yield consistent color balance across frames, but this does not account for differences in lighting (cf. white balance). However, that's of course something we always have to contend with, anyway.

There’s an adjustment layer that lets you do pretty much the same thing with a little bit of extra flexibility all in a single dialogue. It’s called “Selective Color”

There are several ways of accomplishing the same; I think the method proposed here is interesting as one more option in the toolbox. I personally just use curves. If it's a tricky scene, I may use sample points to adjust color in specific parts of the image (notably areas that should render close to neutral grey).

PS: the 'final image' looks to me to have a cyan cast. I personally believe it benefits from a few mild S-shaped curve adjustments, since a simple correction only on the red channel doesn't fix it. Of course, color rendering is highly subjective and there's no single right or wrong result to end up with, but I feel the result presented in #1 is a little muted.

 

[Korbel]

@koraks I use the same settings for the whole roll of film, like exposure and infrared retouching strenght. I have tried it on 3x 120 rolls and on all I could get away by just copy pasting the channel opacities to get a result I liked.

My scan automatically inverts the colors before saving them. In my case the problem was the ICC profile (I found for free on net, no idea how it was made) didnt produce consistent colors at all and the problem was, in my opinion, not white balance, because I got purple tones in outdoor scenes.

As for the other approaches, I think everyone should use the method he is most satisfied with. I know the method for doing white balance via curves, but I didnt find it to suit me.

And yea, the grading is still little bit off I admit. Sometimes I feel like that everytime I look at the same image I see different colors.

As for the method proposed by @_T_ it is quite different workflow in my opinion, because in this adjustment layer you change the hues of the channel and not their relative strenght. I have tried it just now and I cant get the image I want, but maybe I just use it wrong. It is a shame there is no saturation slider for each channel, it could, I guess, then work the same like my method.

 

[koraks]

My scan automatically inverts the colors before saving them.

The way to work around this (and yes, it really does create variation between frames as you'll find out sooner or later) is to scan everything as a positive, then invert & color balance manually. If you do this with the exact same exposure settings, i.e. you manage to disable auto-exposure on the scanner, you can get actual frame-to-frame consistency. Mind you, it's often not possible in the scanning software to disable auto-exposure.

the problem was, in my opinion, not white balance

White balance is a problem/challenge phenomenon that comes on top of this. If you decide on a per-roll consistent workflow, what you'll end up with (if everything works as intended) is a standard color balance across all frames. This will be OK for those frames for which the lighting matches the lighting conditions on the frame(s) you based the default color balance on. For frames with other types of lighting, the balance will be off.
I'm using terms here that suggest there's a right or wrong color balance, which objectively may be the case, but it's usually not the way we work for amateur or fine arts photography, where color balancing is more of a subjective effort.

So essentially you have a couple of color choices stacked up on top of each other:
1: The color balance of the negative resulting from auto-exposure decisions within the scanning software & firmware
2: The color balance resulting from inversion of the negative
3: A default adjustment of color balance to get all frames close to some kind of average neutral rendition
4: Adjustments for frame-to-frame differences in lighting conditions and exposure
5: Adjustments 'to taste' to suit the preferences of the photographer for a particular frame.

We often take several of these steps in a single stride to get where we want to be. This for the most part works fine especially when working on individual frames at a time. When you try to batch-process images and/or try to get a consistent look across a series, it really helps IMO to be aware of which steps affect the color balance and how to control these steps for the purpose you have in mind.

Any process that tries to fix some kind of default/standard balance starting from step 3, while ignoring steps 1 & 2 is doomed to fail one way or another. It's like building a house on top of a lake, using a couple of randomly selected rafts as a foundation. Everything may appear kind of solid, until the wind picks up and your living room floor all of a sudden turns out to be a couple of inches higher than the kitchen floor.

the ICC profile

An ICC profile will generally not work for scanning and color balancing negatives except under very specific, controlled conditions; it's no surprise you had little luck with this. ICC profiles made using an IT8 target (which you should ideally make for your individual scanner, but a downloaded profile for your type of scanner may be close) will help for scanning slides/positives and/or reflective media if your scanner supports them.

 

[Korbel]

And also I think other approaches tend to adjust the hue of each channel and in the end the colors are all over the place. Keeping the channel hue at fixed point and just changing its strenght is the new thing this method brings to the table.

But Im not a photoshop engineer and I have no idea what it does internally. Thats just how it feels when editing to me

 

[Korbel]

Any process that tries to fix some kind of default/standard balance starting from step 3, while ignoring steps 1 & 2 is doomed to fail one way or another.

Yes, thanks for clarifying this up. I should have included this in the original text too

 

[koraks]

Btw, this is what a simple curve adjustment on the original in #1 gives me (left) compared to your end result (right):

It's a single adjustment layer which takes just a few minutes to dial in.

The reason why I generally prefer a curve adjustment on the RGB channels is that more complex adjustments have two drawbacks: (1) they overlap/interact, and they are non-transparently non-linear. Or, put simply: you never quite know what the heck is going on. For instance, your approach of adjusting separately RGB as well as CMY is conceptually odd IMO, since an R adjustment is the inverse of a C adjustment, etc. It doesn't make a whole lot of sense to me to adjust the same channel in multiple ways. This is further complicated by the lack of a good definition of 'cyan' in an RGB system (which you're working with in your image), and the logical conclusion is that your 'cyan' channel is really a combination of the green + blue channels in some (known?unknown?) ratio. This means that your cyan adjustment does not act as the pure inverse of a red adjustment, but probably ends up altering the relationships between the three color curves in an uncontrolled way. A practical drawback is that the more adjustments you pile on top of each other, the bigger the risk of losing something in the process or pulling something grossly out of whack. In my view this already happens in the example image above where your end result incurs a magenta/green crossover, while overall color information is being reduced (lower saturation; everything tends towards grey/brown).