The flipside – A closer look at printing color negatives

In previous blogs, I’ve focused on how color paper responds to a light source. When it comes to printing photos, this leaves out a rather relevant bit – the negative. In this blog, I’m going to explore the interactions between light source, negative and color paper. Hopefully, this will give some more insight into which LEDs work best for color printing (although I have a pretty good idea already…)

To simplify matters a bit, the choice is really only between different flavors of blue and two flavors of green. I’ve already established that red really doesn’t matter, as long as the LEDs used don’t have any emissions that hit the green/magenta layer.

The obvious thing to look at is how the negative transmits or blocks different wavelengths of light. The challenge associated with this, however, is to figure exactly this out. Fortunately, Kodak once more comes to the rescue with spectral dye density data for their color films. Unfortunately, on the other hand, they only split this out per individual dye for the Vision3 motion picture films, and not their C41 still photography color negative films.

This is what the datasheet for Vision3 250D gives in terms of spectral dye densities:

Before we go any further, I feel it needs to be established whether this is representative for C41 films as well. I think we’re in luck, though. Here’s the plot above with the spectral dye densities for a mid-grey subject photographed on Portra 160 overlayed in cyan:

What I’m looking for in the plot above is whether the density peaks match. And that’s looking reasonably well. It seems that the yellow dyes on both films peak out at around 450nm. The magenta dyes also have fairly narrow peaks at around 550nm, although the C41 dye seems to have a much wider peak that extends into lower wavelengths. I assume this plateau belongs to the magenta dye, because of the small dimple at the right of the yellow dye peak that seems to separate yellow from magenta.

The cyan dye peaks seem to be slightly shifted, with the C41 dye peaking out at 700nm and the ECN2 dye peaking slightly earlier at around 685nm. However, since cyan is quite well separated from magenta anyway, it seems that wavelength choice won’t be critical for the red channel in this regard.

Note that the paper doesn’t seem to care much about which red we choose as per my earlier blog, but what’s the situation for blue and green? In that blog, I pit 450nm and 480nm against each other for blue, and 525nm against 550nm for green. From the perspective of the paper, 550nm performed best, with no crossover upon overexposure. On the blue front, results were very inconclusive. In fact, there didn’t seem to be much difference in color rendering.

Let’s have a look at the greens first. I’m going to be using the spectral transmission data I can find for the same or similar LEDs I’m using, and extrapolating from this to fill in the gaps in my data. For instance, for green, I don’t have the actual spectral data for my 550nm LEDs, but I do have them for similar 515nm LEDs, and I simply shifted this spectrum to the right so the peak wavelength hits 550nm. Is this accurate? I don’t know, but given that LED emission peaks are all fairly narrow bell curves, I have a reasonably good feeling about it.

See how the 550nm LED would hit the peak of the magenta dye on Vision3 250D very nicely indeed. More importantly, it doesn’t seem to take in too much of the yellow or cyan dyes’ density curves either. Insofar as these 550nm green LEDs also interact with other dyes beside the magenta one, it’s mostly the cyan dye. An error to either side has its drawbacks; if the cyan dye also blocks the green light used, it means the red image doesn’t only materialize as red in the print, but also as green. In other words, red (and cyan) become contaminated. The corresponding problem happens if the yellow dye (blue image) contributes to magenta dye formation in the print, causing the yellows and blues to foul up.

So, 550nm seems to be a very reasonable compromise for green, and maybe even better than 525nm:

525nm green takes in a little more of the yellow dye on the negative. Whether the net effect in this regard is better or worse than with 550nm isn’t entirely clear. But what was very clear earlier on already, is that 550nm just prints cleaner on the paper, as it didn’t create any crossover into yellow/red upon overexposure. So 550nm definitely seems to be the better choice for green.

What’s the situation for blue, though? I have been printing with the 480nm blue LEDs a bit lately, and to be frank – I have a feeling they might not be the best choice. Something’s going on with the blues and yellows in those prints that’s not quite right. It’s not a huge effect – it’s very much the same feeling I had several years ago when I started testing LEDs for color enlargement, and back then, it was also the blue channel that somehow didn’t feel quite right.

So let’s have a look at how the two wavelengths I compared recently map out onto the ECN2 film’s dye density curves. Starting with the 450nm LEDs, this is how that could look like:

That looks pretty reasonable if you ask me. This wavelength seems to only just miss the yellow peak. Moreover, it doesn’t take in much of the magenta dye image either – although it overlaps with the lower part of its slope.

Here we go with the 480nm LEDs, which in reality look more cyan than blue. Mind you; again I had no emission curve for these particular LEDs, so I shifted the 450nm curve to the right so its peak is at about 480nm.

It’s clear that these LEDs don’t quite hit the yellow dye’s peak. In itself, this is not necessarily all that much of a problem – it just means that the blue channel ‘sees’ a little less density than it optimally could, which could be compensated for by reducing exposure on that channel.

However, what’s worse, is that this flavor of blue light interacts quite significantly with the magenta dye at the same time. Since it also doesn’t quite hit the yellow dye’s density peak, the difference between the yellow and magenta images from the perspective of the blue channel becomes more limited. In practice, this means that the magenta dye (green image) contributes quite significantly to the forming of yellow dye (which should only belong to the blue image). The result would be a rather severe crosstalk between the color channels. Crosstolk between color channels taken to the extreme tends to make everything turn into a grey mush, so this might be the reason why the blue/yellow channel looks a little ‘off’ to me, subjectively, when printing with these 480nm LEDs.

So 450nm would have the edge over 480nm for blue. But what if I err to the other side a little, and go down in wavelength? I recall that when I tried this initially, years ago, the prints also didn’t quite look right. In particular, the yellow patch on a color checker just didn’t render as a nice yellow. As I recall, it seemed dirty and (oh…memory…) it leaned either towards red, or towards cyan (…I forgot…).

Well, let’s ‘fake’ a 425nm LED and see how it would interact with the negative’s dyes:

It’s a compromise to the other side, although I expect it to be less problematic than the 480nm blue. The 425nm LED would be blocked a little by a secondary absorption peak of the cyan dye. I’d say that the 450nm wavelength I was using for years really is a pretty good option all considered.

The above is fairly firm ground, as far as an off-hand analysis can resemble anything firm. What follows next is more speculative, shaky and suspect – so take it mostly for its amusement value.

Since I was mucking about with some dichroic filter transmission curves anyway, I thought I might throw in a little exercise involving those as well. In the musings about which LEDs to use/keep using, I’ve fussed about unwanted absorption by other dyes a bit. How does that situation look with dichroic filters? I’ll fall back on the set of curves I gleaned from Edmund Optics used in my previous blog.

I don’t know how representative these are for any old dichroic color enlarger you’d found out there in the wild (or in your own darkroom), so I’ll try to refrain from drawing firm conclusions. After all, dichroic filters can be made in any number of ways; just compare the set of Edmund Optics curves I will use below to a set of curves I found in a publication on dichroic filters for digital CMOS image capture:

Apologies for the somewhat confusing picture above, but the only thing it intends to illustrate is that dichroic filter curves can vary. I’ve displayed the ‘digital’ dichroic filters above in their inverse color; so the red plot is actually the cyan dichroic filter transmission curve, green is the magenta filter and blue is the yellow filter.

Compare the cyan to the red curves and you’ll see how one filter is far more ‘steep’ (the Edmund Optics one) than the other. The magenta filter for the digital application is a much narrower ‘trench’-shaped pattern than the broader ‘bathtub’ magenta of Edmund Optics. For the yellow filters, there’s a distinct difference in cutoff frequency, with the for-digital filter cutting off at 450-480nm while the Edmund Optics cuts off at a slightly longer wavelength.

So filters are not created equal, and thus, I can’t say that the following exercise of comparing dichroic filter curves with film dye absorption is representative for what happens in an enlarger. It really depends on the filter characteristics in the enlarger used – and for all I know, these may very well differ between enlarger brands and models as well. But perhaps we can see a little in terms of general principles.

Let’s start with the cyan filters, which would have to select the cyan dye image on the negative. Here’s what that looks like, overlayed on my trusty set of Vision3 250D curves:

I feel we can safely ignore the bit of near-IR transmission to the right. What’s more interesting is the cut-off point of the filter on the left side, where the magenta and cyan dye images meet. This particular filter curve seems to include quite a bit of the magenta dye image, and it makes me wonder how ‘pure’ a print could be with this degree of crosstalk.

On the other hand, the vast majority of the image formation will be due to the intended cyan dye. Since the dichroic filter covers such a wide range of wavelengths, it can ‘take in’ virtually the whole cyan dye curve. If it also covers a small part of the adjacent magenta dye curve, this contamination remains comparatively small because the intended absorption is so big.

For the magenta filter, there’s a similar picture:

Note how the magenta filter (this one, at least) covers the magenta dye curve quite nicely. But it also takes in quite significant portions of both the yellow and cyan dye transmission peaks. Mind you, if I use the ‘for digital’ magenta filter transmission curve I showed earlier, it seems to match the magenta dye peak of the film much more closely:

So filter choice makes a difference, obviously. However, what these filters have in common is that they capture a rather broad range of wavelengths, and as a result, they will virtually always cause a degree of crosstalk between the color layers on the negative.

Finally, a yellow dichroic filter interacting with the Vision3 dye absorptions:

Note that this also involves a large degree of crosstalk – which again would be reduced if I used the ‘for digital’ filter curve with its lower cutoff wavelength. Still, some compromise would remain:

Interestingly, this ‘digital’ filter cuts through the dye transmission curves at what seems to be a quite sensible point. It looks more like a happy little accident than anything else, since the ‘digital’ filter curve really had nothing to do with color film photography at all.

The main takeaway from all this, at least for me, is that some kind of compromise is always present. This is due to the narrow spacing and overlap between the dye transmission curves of the film, which makes it difficult to get a ‘clean’ signal regardless of what light source is used. At a theoretically level, it seems that a very narrow-bandwidth light source (e.g. laser) would work best, but it makes the wavelength selection quite critical. This in turn brings the question if the wavelength selection would be the same for all types of color film in existence; after all, the color dyes will vary from one film type to another (especially if you compare films from different eras and manufacturers).

Another thought is that the compromise I spot in the exploration above is most certainly considered by engineers in the photographic industry in the past. Film, filter, paper and enlarger manufacturers were aware of this, and their decisions must have been informed by such analyses. It’s therefore very well possible that controls were built into the film in particular to allow for some degrees of freedom in choice of dichroic filters etc.

Finally, there might be a psychological and aesthetic aspect to this as well. Maybe we simply like a color print with a certain degree of crosstalk built into it, because it brings the hues a little more together. Maybe it visually works fairly well if the shadows in a print tend a bit towards yellow due to crosstalk; it may make the scene more natural us than it is in reality. Hey, it’s a thought. Might be interesting to test this one day – although it would have to be a very, very rainy day…