Recently I had some imagesetter negatives so I could get a feeling for how they work with carbon transfer. Well, first results are in and as expected, there’s a learning curve involved. The keyword here is ‘dot gain’ – it played a lot bigger tricks on me than I had anticipated!
First of all, I should really thank Kees Brandenburg for hinting me of a small print shop in our country that still produces imagesetter negatives at attractive rates. Moreover, they turned out to be very responsive and willing to think along when I inevitably came up with some silly questions. So thank you Kees for suggesting this place, and thank you Bart at SK for being patient with me! Further thanks also go to Erik Wetter for sharing his findings with me and Calvin Grier for offering his advice even though he had absolutely no obligation of doing so.
Alright, so I had this imagesetter service on hand and figured I should give it a try after having concluded many times already that this is probably the best way forward to make consistent carbon transfers from digital files. I cooked up a test pattern/sheet and sent it out. This is the design I had them output for me:
The upper part is a collection of seemingly random patches. This is in fact a 160-patch step chart I generated using ArgyllCMS. So far I’ve been using it mostly as a more fine-grained alternative to the more common steps you can see lower down the page next to the circular gradients. The patches I use for density measurements and are the heart of the matter. I included the gradients to get a visual/subjective impression of the greyscale progression and to spot any major problems with banding, moire or whatever other optical effects might result. For good measure I also included two photographs, again for subjective assessment.
I assembled all this in Scribus, which allows for output to EPS. The reason I chose this approach is that it avoids any problems with gamma or ICC profiles that I might have run into if I had sent TIFF or JPEG to the print shop. While the pictures and the Argyll patches were included as bitmaps, the 0-100 steps I made in Scribus and defined on the basis of fixed K-values (similar to L from Lab*), which means the printer receives precisely what I defined with no risk of anything getting lost in translation.
I had this output as wrong-reading negative, which means that the printshop has done the lateral inversion (mirror image) and density inversion (i.e. black becomes white and vice versa). For the halftoning, there weren’t many options to choose from, so I went with the finest screen that was offered as a default, which was a 180lpi amplitude modulated screen. Here’s the final film output as photographed in my small printing frame:
It’s an imagesetter negative, so it only has two tones: black and clear. The black density I measured to be around 3.9logD or thereabouts. Plenty dense enough for my needs! The halftone pattern is entirely invisible to the naked eye (well, at least my naked eye), but it can easily be studied with a powerful loupe. Prettiest dots I’ve ever seen! Sharply defined and flawless; great job.
Here’s a look at a scan of the halftone screen. Note that I scanned this with my 4990, which will inherently result in a rather soft image. The section is effectively a 1200dpi capture with some sharpening applied. It does no justice to the sharply defined nature of the dots, but serves to illustrate the idea of the amplitude-modulated halftone screen, which is commonly used in e.g. newspapers.
Now, the surprise came when I tried to produce a carbon transfer from the greyscale steps and plotted the densities as measured from the print. Totally useless! Very simply put, all tones came out way, way too dark. Here’s the curve I obtained:
Uhm…no, that doesn’t look right at all! The horizontal axis is the nominal K-value from CMYK that I sent to the imagesetter service. Vertical is the normalized L-value I measured with my i1Pro spectrophotometer.
My first thought was that there was something amiss in the digital part of the process. Specifically, I suspected that the imagesetter service was interpreting the greyscale values differently than I had intended them. But we could swiftly rule this out because (1) I had used an EPS format as indicated before, so the risk of such ‘lost in translation’ problems should have been minimal to begin with, and (2) to be sure I verified with Bart at SK that he received those patches correctly, which he could quickly confirm. Also, if I put the imagesetter negative onto a black sheet of paper, the tonal progression on the steps looks perfectly fine. So what’s going on?
Coincidentally, at pretty much the same time I was approached by Erik Wetter who had been playing with some imagesetter step tablets that he had made together with John Isner. John happened to have discussed those imagesetter negatives on groups.io because he had trouble interpreting transmission density readings on those negatives. In fact, I first saw John’s groups.io thread and then a few days later Erik appeared in my mailbox; I hadn’t made the connection initially, but Erik soon confirmed we were talking about the same set of negatives. But the thing is, Erik had not just been measuring the optical density on the imagesetter negatives – he had also been printing them with carbon transfer, and even with a tissue formulation I had given him a few weeks before. Guess what…he had the same problem of a deeply sagging curve.
If you want to cut a very long story short, then the gist is in Calvin Grier’s swift response to the groups.io discussion I linked to above. He mentioned dot gain and basically, that’s the whole explanation. The question then is, how to deal with it? But let’s do a bit more of the long story first.
Prompted by Erik’s message, I also measured the transmission density on my imagesetter negative. Sure enough, this yields the same kind of dramatically skewed/sagging curve that John Isner reported on groups.io. Here’s mine:
This is a plot of the measured logD on the imagesetter film against the K-value of the steps I sent out to the imagesetter service. Now, this kind of measurement is kind of silly and meaningless, as Calvin also remarked when John posted his question. It’s firstly meaningless because the purpose of an imagesetter negative is to be printed, so the key question is how the printed densities turn out. What densities you measure on the film isn’t very relevant – at least not in the way it is when you’re working with continuous tone negatives (e.g. in-camera silver negatives).
The second major problem is that the fact that we’re now talking about halftone screens is very pertinent. A regular photographic transmission densitometer as used by John and also myself in doing this measurement, assumes an evenly distributed density. A halftone screen is anything except that – it’s literally just opaque dots and blank film in-between. The only meaningful things you can measure with a regular densitometer on an imagesetter negative is the base density of the blank film and the opaque density of its dmax. Everything you might try to measure in-between is just, well, bogus. On a ‘greyscale’ patch, what a regular densitometer sees is just some light seeping between the halftone dots. How much density it’ll register depends on an interaction between the optics of the densitometer and the halftone dot size and only to a limited extent controlled by dot coverage (amplitude or frequency).
It’s different of course if you use a measurement instrument specifically made for measuring halftone screens. That’s the kind of device a print shop would use in their calibration process – which is exactly what Bart at SK does. So he could confirm to me that especially the darker densities are consistently within a few % on target in his process.
So, when working with halftone imagesetter negatives, resist the urge to measure the negatives with a transmission densitometer. Or at least resist the urge to interpret the outcome as meaningful. Thus, let’s get back to the more meaningful approach, which is the printed image. Because we still had a major problem going on there as well.
I considered dot gain to be a problem, as I had been aware of it for a long time already (you’ll find it referenced in several of my earlier writings). But I rejected that it would be the main cause of this problem, because…well, how bad can a dot gain problem be!? Surely, not this bad. So it just has to be something else, right?
I reached out to Calvin and presented the problem to him. I mentioned dot gain couldn’t be the main explanation because (1) the problem was just way too big and (2) the dot quality on my carbon transfers seemed to be pretty good, really. Examining the dots with a powerful loupe just showed very nice, sharply defined dots. But dot gain was still the only concrete issue Calvin could come up with. He generously did offer to have a look at some physical samples, so I set to work on making them – and as is often the case, in the process of trying to systematically describe the problem, I started to understand it.
Since dot gain definitely does play a role (I just thought it would be a marginal one), I started by trying to reduce its impact. Consider this very simple model of dot gain in a process like carbon transfer:
Here, dot gain (represented as a red corona around the theoretically perfect black replica of the original dot in the negative) is the result of the cone of light ‘fanning out’ in the print medium due to its thickness. A thinner medium (center image) will exhibit less dot gain, because there’s less distance for the light cone to fan out into a bigger dot.
A contributing factor is the degree of collimation of the light from the light source. The better the collimation, the less influence a thicker print medium has on dot gain. In the right-side illustration, better collimation is modeled as a larger distance between the negative/print and the light source – which in practice is a fairly effective way to improve collimation without adding any optics or changing the light source itself.
In terms of collimation, I had already invested some resources into an improved light source – which is one reason why I figured dot gain shouldn’t be much or a problem anymore. So I decided to attack the problem from the other side – the media thickness part of the equation. And the simplest way to test that, I figured, was just to print with an essentially zero-thickness medium. How about plain old cyanotype?
So I did some cyanotypes. And found out that…the problem was still there. Drat! But since cyanotype is a quick & easy process, I could accelerate my testing and found out a few notable things:
1: Cyanotype really isn’t a “zero-thickness” medium. There’s always a layer that accepts the sensitizer and especially when using paper, the open structure of the paper surface plays a role in dot gain.
2: Exposure plays a role. Higher-intensity exposure (time, light source intensity) results in more dot gain.
Bearing #2 in mind, I made a family of cyanotype curves at various exposure times, for 500, 100 and 80 (arbitrary) exposure units, respectively:
Above is a plot of the three exposures together with a linear reference (theoretical ideal). I’ve plotted the normalized L-values in the chart above.
I have to admit to being a bad boy and worked in a second parameter – the 100 and 500 unit exposures were made on Schut Salland paper and the 80 unit exposure on Fabriano Artistico, which has a slightly finer surface texture. Either way, the pattern is quite clear: overexposure results in a lot of dot gain as evidenced by the 500-unit exposure.
Now, as indicated, I plotted normalized L-values above, so what you cannot read from this is absolute density. Here’s another plot of the same exposures, but now showing measured logD:
Note that the difference in exposure between the 100- and 500-unit curves was over 2 stops, but the absolute difference in maximum density is only around 0.1logD. This means that the 100 unit exposure was nearly enough to hit dmax for the cyanotype process. Since a halftone printing process basically only prints a single density, there should (theoretically) be no difference between two prints as long as the exposure was sufficient to hit process dmax. The cyanotype example above illustrates that there’s a bit of a tradeoff in real-world printing conditions, at least in the setup I used. An exposure close to the 100 unit curve seems to be like a decent starting point for linearization (if I would want to go there) with good dmax and limited dot gain.
Getting back to the main storyline – the 500 unit cyanotype curve shows clearly that dot gain alone can be responsible for the dramatic deviations I (and Erik Wetter as well) observed in my initial carbon transfer test. I was wrong in initially rejecting dot gain out of hand as a possible dominant factor!
Now, the question is how to deal with it. The first thing is simple – the initial carbon transfer test I did with an exposure intensity that was typical for how I print continuous tone negatives. However, the cyanotype test showed me that exposures for a halftone approach can be significantly shorter. So one factor contributing to the dramatic dot gain in my first test was in fact simple overexposure.
Another factor is in the thickness of the medium. Referring to the dot gain illustration I gave above with the light cones: I could of course optimize the carbon tissue to be better suited for this particular application. My first test was with a tissue I had coded as “XMPD1t”, which describes a tissue using Kremer XSL black pigment, medium pigment load, DAS-sensitized and with a high thickness. To characterize the tissue thickness, I pour this to a wet height of ca. 1.1% and the glop has a 11% w/v gelatin load, resulting in ca. 12.2mg/cm2 of dry gelatin on the tissue. This tissue formulation I developed as an optimization of continuous tone printing, specifically to push back the tonal threshold and thereby improve highlight performance.
As it turns out, the requirements on a DAS carbon tissue for halftone printing are in a way opposite to one for continuous tone negatives. The latter works best in my experience with a thick tissue and a relatively low pigment load. A halftone tissue should be kept as thin as possible to limit dot gain, and this necessitates a much higher pigment load. Effectively, the pigment load in terms of dry pigment per surface area unit may work out as the same, but the pigment-to-gelatin ratio glop recipe will be wildly different.
As a proof of concept, I cooked up a much thinner, highly pigmented carbon tissue. In doing so, I also kept an eye on the set of UltraStable formulations that were shared by recently deceased Charles Berger. They now made perfect sense to me – when I first saw those formulations, they puzzled me as the pigment load seemed absurdly high and the gelatin content surprisingly low to me. I was used to making tissues for continuous tone and the UltraStable tissues went pretty much against the grain of how I had come to understand how things work. But UltraStable was designed specifically for halftone work!
I tested the thin tissue, and here’s what came out:
I’ve plotted the original test with the very thick tissue and long exposure by means of comparison. Note how the thin tissue with a short exposure yields a curve that’s a fairly close match to the ideal. The main problem is a sudden jump from paper white to the first density, which suggests that something odd is still going on in that part of the curve. But the process is far better behaved than it was before. And dot gain seems to have been the sole factor responsible for the difference!
Does that mean everything’s OK now? Well, not quite. This was just an initial test to verify the role of dot gain. There’s some fallout of the relatively extreme tissue formulation which may need some further optimization. Specifically, there seems to be a high degree of pigment staining going on, which seems to be due to the combination of a high pigment load and (I suspect) the dispersant technology used in the XSL pigment. I intend to do some further testing with a DIY pigment dispersion made from plain carbon black to see if I can get that to print a little cleaner. And then it’s a matter of getting my finger behind the odd highlight bump that I’m not entirely sure about, yet.
But so far, this is progress at least. I’m going to conclude with a small scan of the test images on this halftone negative that I just printed on the new, thin tissue without any curve correction whatsoever. Considering everything, I think they come out quite promising and to be frank, if the intended output is a simple monochrome print, I’d feel comfortable just printing like this without any linearization whatsoever. Just straight from GIMP to the imagesetter service!
