Printing from an 11-megapixel file in 2025 sounds like a recipe for soft, pixelated results, but modern AI upscaling has changed what's actually possible. The gap between what a low-resolution file contains and what you can put on a large print is now much smaller than it used to be.
Coming to you from Keith Cooper, this detailed video puts two PermaJet fine art papers through their paces on the Epson ET-8550, but the real substance is a close examination of AI upscaling and what it actually does to low-resolution source files. Cooper uses a 2004 shot from near Crested Butte, Colorado, taken with a Canon EOS-1Ds at 11 megapixels, and pushes it through Topaz Gigapixel AI at six times upscaling to fit an A3+ sheet. The image is dense with bare Aspen trees, fine branches, and distant ridgelines, exactly the kind of content that exposes what upscaling software is actually generating. What comes out looks, in his words, "plausible," and when you see the microscope shots of the actual ink dots on paper, the distinction between real and synthesized detail becomes almost irrelevant at viewing distance.
The more interesting finding is how much the input file shapes the output. Cooper runs the same raw file through two different processing paths: a straight Adobe Camera Raw conversion and one pre-processed through DxO PureRAW, which he also uses regularly on files from his Canon EOS R5 and Fujifilm GFX 100S. The DxO-processed version feeds a noticeably sharper, more contrasty image into the upscaler, and the difference in the distant treeline and rock detail is visible side by side. The lesson is that upscaling software doesn't compensate for a soft input; it amplifies whatever you give it. If you're working with older, lower-resolution files and want to pull large prints from them, the raw processing step deserves as much attention as the upscaling step itself.
Cooper also raises a question worth sitting with: at what point does AI-generated detail cross a line? For upscaling a landscape to fill a larger print, he's comfortable with it. Extending a frame to fill a different crop is a different matter, and he draws a personal line there. He also flags one practical warning about upscaling AI: crowds with distant, poorly resolved faces can produce unsettling results, and it's worth checking those carefully before sending anything to print. Check out the video above for Cooper's full side-by-side comparisons, the microscope footage of the printed ink dots, and his take on where AI upscaling fits into a real print workflow.
No comments yet