Using AI-generated imagery in print: resolution, colour and the CMYK problem

AI image generators are now part of a lot of creative workflows — mood boards, concepts, backgrounds, texture plates, even finished hero imagery. They're brilliant for ideas. They are not, out of the box, built for print.

Two things trip people up: the files are RGB, never CMYK, and the pixel dimensions are smaller than you think.

Here's the short, painless version.

Every AI image comes out RGB

Midjourney, DALL·E, Firefly, Stable Diffusion — they all render to a screen colour space (typically sRGB). There is no "export as CMYK" button, because the models were never trained on ink.

That matters because AI imagery loves exactly the colours CMYK struggles with: glowing neons, electric blues, saturated oranges, luminous skies. Convert those to CMYK late and they flatten. The vivid teal that sold the concept in the deck comes off press looking like a nice pair of sunglasses got involved.

If you're new to why that happens, our RGB vs CMYK guide covers the mechanics.

Practical rule: convert deliberately, early, and look at it. Don't let the printer's RIP do it at the last minute.

Resolution: what you actually get

This is the bigger constraint. Per Midjourney's own documentation on image size and resolution, generated images land around the 1024 × 1024 px mark for a square, with non-square aspect ratios trading width for height at roughly the same total pixel count. Upscaling features push that up by a factor of two or four — useful, but it is interpolated detail, not captured detail.

Translate that to print at 300 dpi:

• 1024 × 1024 px → about 87 × 87 mm. Roughly a beer mat.
• 2048 × 2048 px (2× upscale) → about 173 × 173 mm. A5-ish.
• 4096 × 4096 px (4× upscale) → about 347 × 347 mm. A3-ish, and softening.

So an AI image is fine for a pack panel detail, a spot illustration or a small insert. It is not, unaided, an A1 poster — and it's certainly not a 48-sheet unless you're working at 1:10 scale, where the effective requirement drops dramatically and AI imagery can genuinely work.

Upscaling: what helps, what doesn't

• Model upscalers (the platform's own 2×/4×) — good for organic texture, skin, foliage, skies.
• Dedicated tools (Topaz Gigapixel, Photoshop's Preserve Details / Super Resolution) — often better on edges and man-made structure.
• Regenerating at a wider aspect ratio — usually beats upscaling a square and cropping.

What upscaling will never fix: invented text, mangled hands, impossible product geometry, or a logo that's 90% right. Those get redrawn in vector, not enlarged.

The artefacts that only show up on paper

Screen hides a lot. Ink doesn't:

• Compression mush in flat gradients, which banks up as banding on press.
• Halo edges from upscaling, visible as a light rim against dark backgrounds.
• Noise that reads as texture on screen and as dirt at 300 dpi.
• Over-saturated blacks — AI blacks are often near 0/0/0 RGB, which converts to a 300%+ ink build. See rich black for what to use instead.

A workable print pipeline

1. Generate at the widest aspect ratio you'll need — don't crop down from square if you can avoid it.
2. Upscale before any retouching, not after.
3. Retouch in 16-bit RGB, keeping the file in a working space (Adobe RGB or sRGB).
4. Convert to the destination CMYK profile (FOGRA51/52 for UK/EU coated and uncoated) with Relative Colorimetric + black point compensation.
5. Check total ink coverage against the substrate — 300% litho, 240–280% for large format.
6. Rebuild any type or logos in vector. Always.
7. Pre-flight the final PDF like any other job.

Rights, disclosure and the boring-but-important bit

Before it goes to press, confirm: does your AI tool's licence permit commercial use at your subscription tier? Does the client's brand or legal team require disclosure of AI-generated assets? Does the image inadvertently resemble a real person, a trademarked product or a recognisable artist's style?

None of these are pre-press questions, but all of them stop a job just as effectively as a bad colour space.

In short

AI imagery is RGB, screen-sized, and unaware of ink. Treat it as a source asset, not a print-ready file: generate large, upscale carefully, convert to CMYK on purpose, and pre-flight the result. Do that and it behaves like any other photograph.

---

Want AI-generated imagery prepared properly for press? Get in touch — we handle the upscale, colour conversion and pre-flight so it lands right first time.