this is, bar none, the most work i've ever put into a print - and it's digital, and not even a print of my own photo!
The original negative is a 4x5 shot by @spork on what I believe is TMax 100. Unsurprisingly for 4x5, it has a lot of detail, so I wanted to make sure that all got into the print. 18x24 isn't enormous, but it's likely to be displayed in a place where people can get right up on it, so any reduction in quality is going to be noticeable, and any reduction in quality that's obviously due to digital processing is going to stick out like a sore thumb.
Unfortunately, I don't have the equipment, space, or expertise to do analog 18x24 prints, much less from large format negatives, so that means it's gonna have to get digitized at some point. I've been wanting to build out a high-resolution MF / LF scanning workflow for a while now, so I figured this was a good chance to do it.
I did the initial image capture with my EOS 5DS R and a Sigma 150mm f/2.8 macro lens at 1:1 magnification. This got me 55 images, which total about 3.5 GB and cover each point on the negative about twice. This, unfortunately, turned out to be the easy part.
Lightroom's panorama stitcher, which I have used in the past with no problem for this kind of work, wouldn't even try to stitch that. I was able to get close by stitching columns piecemeal, but it wasn't able to get the whole thing done in the end and the individual results weren't great either, so I switched to Hugin. After spending quite a bit of time figuring out how to use it properly (one mis-inferred control point will ruin your entire afternoon, apparently), I was able to get it to register the images, but actually stacking them to produce an output file wasn't going to happen on my desktop - enblend wasn't happy getting fifty-five 50 megapixel images at once, and I discovered only after waiting hours for it to make a first attempt that it cannot write a TIFF larger than 2 GB.
So, since I didn't want to keep working on a several-hour iteration cycle, I copied the project over to my Dell R730 and started working directly with enblend from the command line. I eventually found success by working in OpenEXR and stitching 16 images at a time to form four groups, and then stitching those four groups to produce a single, enormous OpenEXR file containing the whole negative.
I should point out, at this point, that all the images I'm working with here are completely radioactive to modern computers. Once you start getting over about a hundred megapixels, Windows is likely to have File Explorer completely hang up if you even open the folder the image is in, and macOS will deal with them but it certainly isn't happy about it, even with a 10 Gb/s connection to the server they're stored on.
In this case, the complete negative scanned to 2 gigabytes, compressed, with a resolution of 23,180 x 27,690 - a little over 640 MP. If you try and put six hundred million pixels in the same file in Windows, it turns out that Bill Gates will personally come to your house and punch you in the genitals. Thankfully, however, I have some macOS systems around, so I can keep going.
But it turns out now I have a different problem: what the hell can you edit that with? And I've got to edit it - first of all, it's still a negative, and I still have to crop and rotate it before I'll be able to print it in a way that preserves the composition.
Normally, I work with DSLR-scanned negatives in Adobe Lightroom Classic, so I figured I'd keep doing that. Two problems there: one, Lightroom doesn't directly support OpenEXR (annoying but solvable), and two, it also flat-out refuses to load images larger than 512 MP.
So I went ahead and installed Darktable and kept going. Darktable, to its credit, has no problem with that 640 MP OpenEXR file - it honestly edits that faster than Lightroom edits normal-sized photos, though of course the interface isn't nearly as good. I was able to invert the negative, adjust the contrast, and get it rotated and cropped as necessary - and then I exported it, and the top half of the image was just a complete mess of random pixels.
Yeah. Turns out something in Darktable's "rotate" operator is screwed up, and if you apply a (non-multiple-of-90) rotation to a large enough image and try to export it, it'll shred part of it, which isn't visible until you actually do the export. And, of course, in this case rotating the image was absolutely required for it to look right.
Thankfully, since I was able to preview the rotation in Darktable just fine, I could simply take the rotation value and use ImageMagick to "pre-rotate" the image (which took roughly five minutes), load it back in, apply the same edits, and export. Final ready-to-print image came out at 19,721 x 26,295, for about 520 MP or a hair over 1,000 DPI.
Printing it was, funnily enough, completely uneventful. Load the printer up, hit print, tell it I want 18x24" borderless, and away it goes. The printer in question here tops out at 1,440 DPI so I'm even using all the resolution.
