Most AI image tools hand back a file around 1024 to 2048 pixels on the long side. Let's Enhance, an upscaler vendor, puts the default at 1024 x 1024 for several popular generators, with some reaching 2048. That looks fine on a phone. Put it on a poster and the picture falls apart.
If you need to upscale an AI image for print, the work has two parts: knowing how many pixels the job really needs, and checking what the upscaler did to your picture. This guide covers both, in the order we do them.
How many pixels does a print actually need?
Pixels needed = print size in inches x pixels per inch (PPI). The PPI depends on how far away the viewer stands.
The same vendor guide gives 300 PPI for prints held in the hand, and 150 to 225 PPI for wall posters viewed from a distance. Treat those as common rules of thumb, not a standard. Your printer's spec sheet wins.
Here is the math for a 24 x 36 inch poster, using the guide's own table:
- 300 PPI: 7200 x 10800 px
- 200 PPI: 4800 x 7200 px
- 150 PPI: 3600 x 5400 px
Now compare that with what you have. A 2K AI image of 2496 x 1664 px prints at about 8.3 x 5.5 inches at 300 PPI, or about 16.6 x 11.1 inches at 150 PPI. A postcard works. A poster needs a lot more pixels than the generator gave you.
How do you upscale an AI image for print without losing detail?
A classic resize stretches the pixels you have. An AI upscaler adds new ones. Vendors describe this as reconstructing detail. In plain words, the software makes an educated guess about what fine detail should be there.
That guess is often good. It is still a guess, and it can be wrong in ways that matter for print:
- Skin and fabric can turn waxy or over-sharpened.
- Edges can pick up thin halos.
- Small text, labels and logos can change shape, because the model does not know what they said.
- Repeating patterns, such as bricks or woven cloth, can drift out of line.
We have used AI imagery in production for about the last year, and these are the problems we see most. An upscaler is a tool for adding resolution, not a way to recover facts that were never in the file.
What do we check before a file goes to the printer?
We look at the upscaled file at 100% zoom, in sections, not just fitted to the screen. That is also what the vendor guide recommends. Our list:
- Edges. Follow the outline of the main subject. Look for halos, jagged steps and cut-out fringes.
- Text and labels. Compare every word with the approved artwork. If a word changed, rebuild that area from the real file.
- Texture. Check skin, fabric, paper and metal for a plastic or smeared look. Add back fine grain if the surface went too clean.
- Gradients and skies. Look for banding, the stripes that show up in smooth color transitions, especially in dark areas.
- Color. Confirm the color space and ask the printer for its preferred profile. Smooth blues and saturated oranges can shift when converted to CMYK, the ink set used in print.
- A physical proof. For anything large or expensive, order a small proof at real size. A screen never shows you how a print reads at arm's length.
When is upscaling the wrong answer?
Sometimes the better move is to go back one step. If the image has to carry fine, exact detail, such as a pack label or a product edge, an upscaler cannot be trusted with it. We rebuild those parts from the real artwork or from a 3D model, then composite them over the upscaled picture.
Viewing distance also helps. A billboard seen from the road needs far fewer pixels per inch than a magazine page, so a moderate upscale can be enough for large formats read from far away.
Should you upscale once, or in several steps?
Once, where you can. Every pass of an upscaler adds its own guesses on top of the last pass, and small errors stack up. A single 2x or 4x pass from the cleanest source you have usually looks more natural than three small passes in a row.
Start from the best file, not a screenshot or a compressed copy. If the generator offers a higher native size, use it. Pixels that came from the model itself are better than pixels added afterward, and the vendor guide makes the same point: for a new image, a higher native resolution can remove the need to upscale at all.
Also finish the creative edits first. Retouching, cropping and color grading are easier on the final-size file than redoing an upscale after every change. Our own order is: approve the image, fix the content, upscale, then inspect and sharpen lightly. Sharpening is last, because sharpening before an upscale makes the upscaler exaggerate it.
What should you send the printer?
Send the file at the real print size or larger, saved without extra compression, in a format your printer accepts. Tell them the viewing distance and the material, since a matte paper, a canvas and a backlit panel all show detail differently. Ask for a proof if there is any doubt, and say plainly that the image began as AI output so nobody is surprised by soft fine detail.
Practical takeaway
Before you order a print from an AI image:
- Work out the pixel size from the print size and the viewing distance.
- Upscale in one clean step, not several stacked ones.
- Inspect edges, text, texture and gradients at 100%.
- Rebuild anything that must be exact.
- Proof at real size when the job is large.
That is the finishing pass we run on AI images before they leave the studio. If you want to see how we handle post-production for print and digital work, have a look at our portfolio at 35milimetre.com.