Published May 12, 2026 · By CampaignsLive · Production
There is a moment that recurs in every team’s first contact with AI imagery at production scale. The creative looks great on a laptop. It looks great on the agency’s internal Figma. The print partner takes the file, prepares the prepress, and sends back a proof. The proof looks like a different image.
The gap between what AI image tools advertise as “high resolution” and what print production actually needs is wide enough to drive a campaign budget through. This post is about what closing that gap actually involves.
What “high resolution” tends to mean in AI image tools
The defaults across the AI image market settle somewhere between two and four megapixels. That is enough resolution to look excellent on a Retina display, an Instagram feed, or a YouTube companion banner. It is not enough resolution for almost any print placement.
The math is straightforward. Print is typically reproduced at 300 pixels per inch. A 2-megapixel image is roughly 1600×1200 pixels — about 5×4 inches at 300 PPI, which is roughly half a page in a magazine. A 4-megapixel image gets you to a single A4 page. Neither one reaches “double-page spread” or “6-sheet billboard.” Both will upscale to those dimensions if you ask the tool to, but upscaling does not invent detail; it interpolates it, with consequences that prepress proofing makes immediately visible.
Production-resolution print starts at 12 megapixels and is comfortable at 16. That is roughly the resolution of a current-generation DSLR sensor, which is not a coincidence — a great deal of brand creative is shot at that resolution today.
What 16 megapixels buys you in practice
At 16 megapixels — about 4900×3300 pixels — the same source file is enough for:
- A magazine double-page spread at native 300 PPI without upscaling.
- A 6-sheet poster (about 1.2m × 1.8m) at full resolution.
- A 48-sheet roadside billboard scaled at print-spec PPI, which for OOH is lower than press print because viewing distance is greater.
- Bus shelters, transit panels, station dominations, and retail point-of-sale at full bleed.
- An editorial cover where every detail in skin, fabric, or environment is studied closely.
Larger formats — wallscapes, environmental, airport dominations — start to ask for upscaling, but starting from 16MP means the upscaling is a small step rather than a leap. The difference between “interpolate from 4MP to 50MP” and “interpolate from 16MP to 50MP” shows up in the proof.
Why CMYK is harder than it looks
Almost all AI imagery is generated and delivered in sRGB. This is reasonable for screen output and is what most downstream tools expect. It is also where print production starts losing the file.
sRGB has a wider gamut than CMYK. The most saturated, deeply chromatic regions of an sRGB image — the ones that look most striking on a laptop — fall outside what offset and digital print can reproduce. When prepress converts the file, those colors collapse. The collapse is rarely uniform, which is why the proof looks like a different image: the relationships between colors have shifted.
The right response is not to generate in CMYK from the start (which most image models cannot do anyway) but to handle the conversion deliberately. That means:
- Working with the printer’s actual ICC profile, not a generic CMYK.
- Watching saturated regions during conversion and adjusting source-side rather than trying to fix it in prepress.
- Using rich black for shadow regions instead of 100K, so deep tones hold against ink limits.
- Pulling highlights away from the printable maximum so they survive flat-tint reproduction.
- Embedding color targets and reference patches on press-side calibration jobs.
None of these are exotic moves. They are what production designers do as a matter of routine. They are also what almost no AI image tool delivers by default, which is why the file that looked great on a laptop did not survive the trip to press.
Bleed, crop marks, and total area coverage
Beyond resolution and color, three production fundamentals trip up generative output regularly.
Bleed. Print files need image content extending past the trim edge by 3–5 millimeters so the cut does not leave a white edge. Most AI image tools generate to the edge of the canvas with no overshoot. The fix is composing the output to leave room for bleed or generating with explicit bleed allowance.
Crop marks. Standard prepress workflows expect crop marks at the trim corners, color bars on the press sheet, and registration marks for multi-plate presses. These are not generated by the image model; they are added by the production workflow. The file the model produces has to be hand-offable to that workflow without retouching.
Total area coverage. Most coated stocks have a TAC limit around 280–320%. An image that combines 100% cyan with 100% magenta and 100% yellow in shadow regions exceeds that limit and will smear or set off on press. The image looks fine on screen. It fails on press. Production designers compose to avoid this. AI image tools mostly do not.
What “production-resolution by default” actually means
When CampaignsLive describes images as production-resolution, the underlying claim is that the output is meant to leave the platform and be handed off to a printer or media partner without an intermediate retouch cycle. That involves the resolution piece, but it also involves all of the above — gamut-safe conversion, ICC-correct color, bleed, total area coverage, the unglamorous parts of getting a creative file across the threshold into a printable file.
It is the unglamorous parts that make the difference visible in the proof. The visible part — the image on the laptop — is the easy part.
For the full output workflow and the placements covered, see the Print and Out-of-Home use case page.