Sickly colors – the crossover issue

I like to print color RA4. That’s no secret. And I like to tinker. I’m not alone in this, and as a result, working with color film and printing color RA4 pops up regularly in one way or another on the photo forums. One issue that often meshes into technical discussions on ‘analog color’ is that of color crossover. Some people immediately ‘get it’, probably because they have experience looking at and analyzing color crossover. Others struggle with the topic and don’t know exactly what to look for. In this post, I’m going to try and illustrate the issue through some (digital!) mockups.

Disclaimer: in the color domain, this has involved my work with color LEDs for RA4 printing, but also trying to get Kodak Vision3 to work for RA4 printing, DIY-ing my own RA4 developer, developing RA4 at lower temperatures, monitoring RA4 developer health & pH, messing about in many ways with C41 developer (push/pull, lower temperature/extended time etc.) as well as ECN2 (push/pull, modify pH/change development time) – and of course I’ve had to do my fair share of scanning (positive + negative film, prints, artwork etc.) and digital color correction/restauration etc. It’s just too much to sum up, but if I add it all up, I can say that I’ve spent a fair amount of time working with color, looking at it critically and to an extent analyzing it. What I am not, however, is a professional colorist and I do not have any formal education in a relevant domain pertaining to color. So this here is my best shot on the subject as an amateur who has bumped his head more often than he could count (and will continue to do so).

In this post, I will assume a familiarity with the basics of color theory. So if you know what a color wheel is, what the main difference between additive and subtractive color schemes is, you’re familiar with additive and subtractive color and with RGB and CMY(k) schemes, and you know what complementary colors are, you should have no problems following my argumentation and interpreting the examples below. Awareness of color curve adjustments is also essential. Furthermore, it helps if you know the basics of color filter corrections when printing color in the darkroom, but this is not really required to get some use out of this post.

First, what is crossover actually? The term derives from a distortion in the relation between the color sensitivity curves of a photographic material such as color film. That’s quite a mouthful. What does it mean? First, let’s have a look at the color sensitivity curves of a real-world film, in this case Kodak Portra 160:

I colorized the curves above for convenience’s sake. What the chart shows, is how much density the film builds (vertical axis) depending on how much light it is exposed to (horizontal curve). There are differences in the absolute sensitivity (how much density is created for how much light received) between the color curves. This is intentional and corresponds to how color RA4 paper in turn responds to the resulting negative. Well, if you’re nitpicky, you could argue that the difference in sensitivity was perhaps not entirely intentional in the beginnings of color film, but more of a technical limitation film manufacturers had to deal with. But let’s assume for now it’s intentional. It’s also non-consequential to my argument, so why not do away with this difference and shift the red up a bit and blue down a bit to see how the curves relate to each other:

What we see now is that the red, green and blue curves track fairly nicely: they go up in more or less the same angle. There is a bit of a deviation, which again is intentional at least in the center-part of the plot and again relates to what RA4 paper ‘expects to see’ (RA4 paper, I hope, is a non-sentient being, otherwise I’d feel bad about cutting it up) when the negative is being printed.

Now, crossover is if the slopes of the curves no longer track with each other: one slope becomes a little steeper or another becomes a little less steep. Something like this, albeit exaggerated to an extreme extent:

If you were to put into words what happens if you print this, is that in the shadows (the left of the plot) the color balance will be entirely different from the color balance in the highlights (right side). Consequently, if you were to try and adjust this by for instance removing a little yellow from the printing filter, you might fix one end of the scale (let’s say the shadows), but the problems only get worse at the other end (highlights). The reason for this is that a filter adjustment won’t just lift or drop one end of the curve, but affect it in a similar way all over its length. So adjusting the yellow filter won’t fix the blue curve’s slope problem – it will just move the entire curve up or down a bit. So that’s crossover, and that’s also how I often notice it: the colors just look wrong, they look wrong in a different way in the shadows and the highlights, and no matter what you try in terms of filtration, the result seems to only get worse.

To illustrate this, I shot this little scene with my Canon EOS7D (digital is just so convenient for doing a quick illustration!) Shot in RAW, color temperature balanced in post processing for ‘shade’, which matches the indirect natural sunlight situation under which this was shot. All processing was done in 16 bit color resolution in GIMP, with the final results being scaled down to 8 bit for web display. This is the original, which matches quite accurately at least on my monitor what I saw in reality:

I chose the color checker because it’s the very best one I have (coincidentally also the only one), and I added the flowers, foliage and tomatoes to add some saturated real-world colors to the scene. The color checker is nice, but its gamut is necessarily limited to what the offset screen print could manage [see addenda & errata towards the end]. Tomatoes and nasturtiums don’t have this limitation, and having some hues in the scene that are pretty far to the extremes of the gamut helps to spot color problems. These ‘extreme’ hues tend to suffer quite badly if something goes wrong color-wise. Besides, it was a nice excuse to pick a small bouquet from our back yard and enjoy it in the process of writing this post. It’s sitting on my desk as we speak, err, write.

The example image above has no curve adjustments applied to it. This is how a neutral curve looks like: it’s just a diagonal, nothing fancy.

I’m going to use curve adjustments to illustrate the various things that can happen to an image color-wise. I have prepared a couple of adjustments and gave them names as follows:

First, the ‘lift‘ and ‘drop‘ adjustment. These lift or drop the entire curve upward by an equal amount. The curve remains linear and retains the same slope. I’ve shown only the manipulation on the green channel above (green line), with the other channels (red and blue; together they form the purple diagonal) remaining unadjusted, serving as a reference to an unmanipulated curve. In the examples later on I have applied all these adjustments not just to one color channel, but all three.

I started with ‘lift’ and ‘drop’ because I feel these approximate a dichroic filter adjustment when printing color RA4 in the darkroom. Lift and drop do not constitute a crossover problem because the slope or steepness of the curve is not changed: the adjusted curve (here green) remains parallel to the blue and red ones.

This is different in a true crossover scenario, which can be made like this:

These adjustments I call ‘crossover’. As you can see, these adjustments now revolve around a change of the slope of the green curve. Evidently, this can also be done in two ways: on the left, the slope of the green channel is reduced by lifting the shadow-end upward and pressing the highlight-end downward. I would call this ‘crossover green to magenta’ because green becomes dominant on the left-hand side (shadows) and magenta becomes dominant on the right-hand side (highlights). The opposite is ‘crossover magenta to green’, which drops the bottom of the curve while raising the top, thereby increasing the slope of the curve.

The curve remains linear (so not really a curve, but OK) in both cases, but it no longer tracks parallel to the red and blue curves. Remember the extreme example I created based on the Portra 160 curves above? This is basically the same, but in a very clean and linear way. Remember, we could do this on the red and the blue channels just the same.

Note also that by now, we’re not just talking about red, green and blue anymore, but we also encounter the complementary colors: by adjusting the green curve in a crossover fashion, we introduce a green shift on one end and a magenta shift on the other end. Consequently, if we play the same trick on the blue channel, we’ll encounter a yellow shift on one end and a blue one on the other, and with the red channel it’ll be a cyan shift combined with a red shift.

Mind you, this crossover scenario is basically the same as another pair you could make:

I call this pair ‘boost’ and ‘chop’ because the top-end of the curve is lifted up respectively pressed down, while having the bottom-end remaining in the same place. This evidently adjusts the slope of the curve, and hence, it creates crossover. Actually, it’s just a combination of lift/drop and crossover. Therefore, I’ll leave boost and chop for what they are, at least for now, as they’re simply combinations of something we already know.

There’s a final pair I’d like to introduce you to:

I call the pair above ‘more’ and ‘less’. I picked the midpoint of the curve and moved it away from the linear curve left or right, to make a curvilinear adjustment. What this really does, is it alters the actual contrast in that single color channel, and this contrast is furthermore altered in a different way in the shadows vs. the highlights. It’s quite a complex adjustment if you think about it, but a fairly common one to make when digitally correcting images (which I try to steer away from as much as possible for the simple reason I don’t particularly enjoy it…)

Because in more and less adjustments the curve slope actually changes and therefore also loses tracking with the unadjusted red and blue curves (except for an infinitely small spot in the middle of the curve!), it’s also a (complex) crossover situation.

Alright, how does all of this pan out? We’ve got an example image and we’ve got a couple of bullets to fire on it, so let’s have a go. Keep in mind, the adjustments I make below are pretty extreme. If you’re dealing with for instance a crossover situation because you’re using film that’s a few years out of date, the result will likely be much more subtle. I’ve exaggerated the effects below so they’re nice and blatantly visible.

Let’s start with lift and drop. again, I think these best approximate real-world dichroic filter adjustments, so what you see here is something you could pretty much make in the darkroom by printing a so-called ‘ring-around’ (or ringaround):

Note in the examples above that in each case, the effect is a rather even color cast all across the image. Because I did the adjustment in both directions on all three color channels, we get six distinct casts (left to right, top to bottom): green, magenta, blue, yellow, red and cyan. If you encounter this in an RA4 test strip you’ve just made, don’t worry, adjust the appropriate filter and give it another go. You’ll get out of the woods unharmed.

This is not the case if you’re dealing with actual crossover. Again, there are six permutations, but this time I’ve split them into two groups. It’s still the same basic approach as before, just a different adjustment. In both images below, I used the ‘crossover’ adjustments on all three color channels:

Note that things get freaky here. The middle one is in many was the least problematic because it somehow feels kind of natural. The shadows (check the black and dark grey patches on the color checker) tend towards blue, whereas the highlights go yellow (white and light grey patches). This is similar to what a scene might look like with a sunlit and a shaded part, and therefore, I guess our brains tend to accept it as pretty normal. It’s different for the top and bottom rows: here, things start to feel unnatural, at least to me. Again, the grey patches tell the story in the most literal sense, with the top image showing green shadows and magenta highlights, and the bottom image showing reddish shadows and cyan highlights. Particularly the latter I would describe as a ‘sickly look’. Imagine a portrait of a person with Caucasian skin color exhibiting this effect; we might feel they are in fact ill or not feeling quite right.

Above, again three examples of crossover and a similar (albeit inverted) pattern: the bottom image with the red highlights and cyan shadows might not feel particularly problematic, especially if the effect were less pronounced. A little warm/reddish highlights; it’s the kind of thing we associate with sunsets and the ‘magic hour’ just before sunset. But the top and middle rows don’t make much visual sense; I’d call the top image pretty much as ‘sickly’ as the red/cyan example shown earlier (but worse, to my eye!) and the middle one also doesn’t sit quite right.

In all the crossover examples above, keep in mind that a simple filter adjustment when printing using an enlarger won’t help. It’ll just shift one color curve up or down, without fixing the issue. Effectively, you would run into a boost or chop situation – you know, the curve pair we discarded before because they didn’t add anything theoretically new. Here’s an example of all boost and chop for all three color channels:

The chart above is nearly a recreation of the crossover examples we’ve seen before. The only real difference is one of luminosity, but we still see the same kind of skewed color relations. The situation might look less problematic because the absolute deepest shadows will be neutral. The black patch on the color checker comes fairly close, but of course is still lifted somewhat from the bottom of the curve (it’s not a pure pitch black) and hence, it’s still skewed a bit, but not as badly as the highlights.

This is the kind of result you end up with if you try and filter your way out of a crossed-over negative when printing RA4. You could go all kinds of ways, none of them ending up a truthful color balance. It becomes a matter of compromise and selecting the option that feels the least bad of them all. Unless you’re purposefully looking for a crossed-over look, of course, which in itself is a perfectly legitimate thing to do. I would argue, however, it’s nice to have the option of crossover, but not to have to live with it without the option of a neutral balance.

Mind you, it’s just almost as difficult to get a crossed-over RA4 print from a non-crossed-over negative as the other way around. It’s easy to induce a color shift, not so easy to induce a crossover. You really have to start twisting process parameter knobs such as RA4 developer pH, development time etc., or add things like sulfite or peroxide to the developer – measures sometimes argued as being effective contrast-adjustment tools, but I personally think they always come at the cost of at least some color crossover. But that’s another matter.

The illustrations above I hope give some substance to the term ‘crossover’ which I confess I have used many times without much further explanation. Still, it’s all kind of synthetic, perhaps. So how about some real-life examples? Being of the experimental type, I have a few here and there:

Above is a scan of an RA4 print onto Fuji Crystal Archive Supreme paper from an ECN-2 developed Kodak Vision3 50D negative. The developer pH was adjusted to 10.50 (it should be 10.25 officially). Note the cyan-blue shadows and the yellow highlights, suggesting a complex crossover situation involving the red/cyan as well as the blue/yellow axis. While technically uhm…not so good (I’d say, pretty darn horrendous), this example also illustrates that crossover might sort of work. In a way, it’s not necessarily an ugly color balance, although taste is of course a personal matter. Going back to what I said earlier, a rendition with cool (cyan, blue) shadows and warm (yellow) highlights feels sort of natural to us and thereby is somewhat acceptable. Although it’s pretty extreme here, wouldn’t you say?

Another example from the same era; I mentioned at the start that I experimented with Vision3 film for a while, and this is from that time. I was gifted some Cinestill 800T (Kodak Vision3 500T sans remjet) by a friend and shot this frame, using the modeling lights of my strobe kit. Again, the color balance is not necessarily problematic from an aesthetic viewpoint – I actually exploited it here for its effect. But in a technical sense, it’s atrocious. Don’t believe me? Well, what if I tell you that the purple background in reality was a white wall just to the back of the model? In the print (Fuji Crystal Archive, optical enlargement), a pretty rainbow is rendered in the shadows involving hues ranging from something blue-ish to something close to maroon, while the highlights are a golden yellow. You can just make out the blue-cyan brim of the hat; this is in reality near-perfectly neutral grey felt.

Right, one more, but far more subtle this time. Again, the film is from the Vision3 family, back to 50D this time, processed ECN-2 and printed onto Crystal Archive II paper:

Here, you might say that the colors are almost natural – is there anything wrong at all with this image? Well, there is if you know where to look – and if you’re familiar with the real-world scene, which I happen to be. The courgette exhibits a fairly normal hue and color saturation, a bit to the neutral side I’d say. The bell pepper, however, does look a rather neutral red, but very much lacking in saturation. If you look closely, the red in fact isn’t entirely clean and probably contaminated a bit with cyan and/or blue. Blue is likewise problematic; the glazing on the cup almost seems black & white in this print, but in reality, the pattern was a deep and very saturated blue. So here, what happens is crossover resulting in a compromised print where saturated colors (far off-center in the gamut) suffering quite badly, while more central hues in the gamut aren’t evidently problematic. There’s also a more typical shadow/highlight crossover issue going on, but it’s less conspicuous here than in the preceding examples.

The main reason I showed the examples above is a rather prosaic one: my interest is more in making physical prints than in doing digital work, and as a result, I don’t scan much. I made the Vision3 scans because I was in a process of trying to get that film to work for me, which involved posting some examples online here and there and also conversing about the issue briefly with (may he rest in peace) Ron Mowery/Photo Engineer.

Stashed away in boxes and also in landfills in this country, there are many more examples of where I ran into crossover situations in my own testing and printing. Below are some of the issues I’ve encountered along the way. Make of this list what you will – exploit its artistic effects, avoid mistakes I made or prove me wrong.

• Running C41 processing at non-standard temperatures. You may or may not get away with a small deviation as long as you correct for this by adjusting development time, but at some point, crossover will bite. Initially, it will be mostly a shift as opposed to crossover, but the latter is inevitable at some stage.
• Reusing C41 developer without replenishment. Same; this will work to an extent, but at some point, the developer drifts too far from its specified state (particularly in terms of CD4 oxidation and pH, I suspect) and negatives will cross.
• Diluting C41 developer. Again, assuming correction by prolonging development, there will be a heavy color shift but ultimately also crossover.
• Taking developer (C41, ECN2) pH too far out of bounds. Although my experimentation with ECN2 and adjusted pH showed mostly color shifts and not so much crossover, at some point, crossover will also occur.
• All of the above to a lesser extent also goes for RA4 print processing. Here, the margins in my experience are far wider, which mostly (I assume) has to do with the fact that RA4 development is effectively to completion (within reasonable limits). However, running RA4 too hot and at too high pH will give massive problems (and not just crossover – you’ll have to do without any white on your paper, even in the unexposed margins).
• Use of an inappropriate light source or faded/wrong filters for C41 to RA4 enlargement. See my blog on RGB LEDs for more info on the former. Concerning the latter: dichroic filters in enlargers are typically quite durable. It’s something I wouldn’t worry about too much, at least not as an amateur printer.
• Trying to optically print ECN-2 negatives (Vision3, Cinestill) onto RA4 paper. I tried just about every trick I could think of, but crossover was always part of the deal. See examples above. I know Cinestill argue that they have (part of?) a fix for the problem with their developer kit, but I am somewhat skeptical about this and the evidence and argumentation they offer to this effect. I know some people get decent results printing ECN2 to RA4, but these are exceptions to the overarching rule of horribly crossed-over examples that I’ve seen, and I’m not sure if it works for every lighting condition (I fear not). Scanning is a better idea and can yield OK results (but again, examples of not-so-OK results are plentiful).
• Using out-of-date color film. This is a bit of a mixed bag; sometimes you’re OK shooting 7-year expired Superia 200 and encounter just a bit of shift, and sometimes you’re in a veritable mess shooting 5-year old Gold 200. I’m not saying Fuji beats Kodak – I’m saying that expired film is really a gamble. If stored well (refrigerated or even frozen), color print film may last a long time. I’ve shot Superia 200 that went out of date some 15 years ago but that was refrigerated all the time, and it’s mostly OK in the crossover department, apart from fairly heavy fog and color shift – but these we can correct for.

Back to the viewing part for a bit. In case you’re working with color materials (film, color paper) and you run into crossover issues from time to time, or you think you might, here are the things that I generally look for when trying to figure out if there’s a crossover issue going on:

• Most evidently, look at a neutral grey gradient that goes from light to dark. If you know it should all be (fairly) neutral, but you see distinct hue differences between light and dark areas of the same subject matter, it’s worthwhile to have a closer look. Especially if the hues involved are green/magenta or red/cyan, with blue/yellow being a little less conspicuous because they may simply indicate a difference in lighting (sunlit vs. shade). While you might think such neutral gradients aren’t all that common, consider that in an urban environment, there’s a generally good availability of fairly neutral concrete, stonework etc. In nature, rock formations, dunes and beaches may offer similar opportunities.
• Evaluate saturated colors – especially if they aren’t as saturated as they really should be. In particular, look for saturated hues that are quite far apart from each other in the color wheel. If one/some of them become(s) more problematic (less pure, less saturated) than (an)other(s), it’s likely due to a crossover issue.
• Skin tones. I haven’t given much in the way of examples of these, but skin tones, particularly if you have skin under varying intensities of exposure, will show crossover rather easily. This is not so much because there’s anything very particular about skin tones; they’re just a couple of hue families that aren’t very pronounced. The particular aspect is in our brain, I think, as we’re generally rather sensitive to anything that looks odd about the skin of our fellow human beings (or our own skin, for that matter). Evolution, I suppose.
• Objective references. By these, I mean things that you just know should have a certain hue – or not have certain hues. A treeless, sunlit snowy landscape may very well exhibit all kinds of blues and yellows, but is less likely to actually contain much in the way of greens and magentas. If you spot green/magenta rainbows in your snow shots, you might be onto something. Skies usually don’t turn cyan all that much (although they can, a little bit, depending on lighting conditions), so a cyan sky is another indication. Clouds in broad daylight (overcast, or the kind that sits conveniently in a blue expanse), for that matter, are always a fairly good benchmark – although interpretation and taking into account real-world phenomena are crucial.
• The ultimate objective reference is of course a good color checker / card. Since crossover is so intimately linked with the degree of exposure, you’d be best off shooting a color checker at normal exposure, as well as underexposed and overexposed. A color checker only has so much reflective density and this is dwarfed by the density range a typical real-world scene will contain. So to make it easier to spot issues, try exposing a color checker at e.g. -1 stop underexposure, 0 stops and +2 stops overexposure or so.
• Learn to trust your gut feeling / instinct. This may sound odd, but speaking only for myself, I mostly spot crossover issues because a crossed over image just doesn’t ‘feel’ right. I don’t think we’re born with this, apart from perhaps the evolutionary aspect (see above), but it’s something that I think just happens to you once as you spend a lot of time looking at color. Usually, I first respond emotionally to color and only then try to figure out rationally what’s going on. Why have all these neurological antennae and not use them, right?

Please allow me to make a few final remarks before I let you go make beautiful things. Color vision is an incredibly complex and broad topic. I realize very well that I don’t even know the first thing about it! But I do know a couple of things that are important to keep in the back of your head:

• Color vision isn’t distributed evenly across the human population. Red/green color blindness occurs in roughly 8% of the male population (but only 0.5% of the female population), and then there’s myriad other ways in which color vision can be objectively ‘off’. If you just don’t see the whole crossover issue, perhaps it’s because you’re not equipped to see it! There are all sorts of color blindness tests online you could do to check if you fall in one of the various color-vision deficiency categories.
• Even among the vast majority of people who aren’t color blind, there is huge variation in how sensitive we are to color, and as a result, how capable we are of distinguishing them. Again, several online tests are available for your amusement. I recommend one of the free versions of the Farnsworth Munsell 100 hues test, which gives you a ballpark estimate of how well you pitch against your fellow humans.
• Apart from the more objective aspects above, color has a huge psychological and by extension subjective dimension. In reading the above, I think you got a clear idea of what my preferences are: color rendition with good fidelity, saturated primaries when they are available, and I’m sort of OK with blue/yellow crossover but certainly not with green/magenta. It may be totally different for you, and you may (literally!) feel entirely different about the whole matter.
• There’s a linguistic limitation that we just have to accept – no matter how hard we try, it’s very difficult (I’d say, impossible) to verify if you are actually seeing the same thing I’m seeing. I can call a cyan shift green and you might call it the opposite, while in reality our brains are processing the same signals in the same fashion, giving us the same kind of emotional response. Are we in agreement, or not entirely? Of course, this matter was a big part in why I wrote this article in the first place, and why I spent quite some time making those digital mockups – because it gets us at least a little bit past this language thing. But it’s still there.

Well, there it is. It all turned out a bit longer than I had envisioned, but I think it’s (1) an interesting topic and (2) it just doesn’t allow for a very brief explanation as too much would get lost in conciseness. Either way, I hope you took something useful away from it. Please let me know if you did (or didn’t)!

Addenda & errata

• In relation to the color checker card I used, it was pointed out to me by @AgX from Photrio that this is likely not offset printed but rather a screen printed card. Closer examination of the imperfections of the chart seems to suggest he’s right at this.
• @DREW WILEY makes a useful addition concerning crossover on yesteryear’s Cibachrome materials, which apparently crossed over on the green/magenta axis as they matured/got older (and ultimately expired). RA4 paper does not do this, but turns yellow with age (including the whites, in my experience). I would add personally that Kodak Endura not only turns yellow, but also green (so effectively a lime color) as it ages. These shifts mostly seem actual shifts and not necessarily crossover. They do render the paper pretty much useless in my experience, so it doesn’t really matter…(although I do have some framed prints on my wall of Ektar negatives printed onto Endura paper that was expired and did no longer have pure whites; I still like these prints.)
• @Mick Fagan also on Photrio adds that as RA4 developer expires, it makes the blacks on an RA4 print turn blue without ever reaching black. Since the effect is density-dependent, I guess you could see it as a kind of crossover. If you get RA4 prints with blue blacks and no amount of exposure or filtering can give you a neutral and deep black, verify if your RA4 developer is sufficiently replenished, not oxidized and its pH checks correctly.