Making AI art that prints well

An image can look excellent on screen and disappointing on paper. Screens are small, backlit and forgiving; a 24×36 print is none of those things. These are the things worth getting right before you list a piece for sale.

Choose the aspect ratio before you generate

This is the mistake that costs the most credits. If you generate square and then decide to sell 16×20, you have to crop, and cropping throws away composition you paid for. Decide the shape first.

If you intend to offer one artwork in several shapes, generate the widest version and crop inwards, rather than generating tight and trying to extend.

Leave room at the edges

Framing eats the outer few percent of an image, and canvas wrapping eats considerably more — the wrapped edge of a stretched canvas can consume an inch on every side. If your subject runs right to the border, it will be cut.

Prompt for it. Phrases like "generous negative space around the subject" or "wide framing, subject centred with clear margin" cost nothing and save the piece.

Detail that survives enlargement

Judge a candidate at full size, not as a thumbnail. Open the image and look closely at:

• Faces and hands — still the most common failure, and the first thing a buyer's eye goes to.
• Text and signage — unless you used Ideogram or GPT Image, assume any lettering is nonsense and crop it out.
• Repeating structures — railings, windows, fence posts, roof tiles. Models drift, and a fence that changes rhythm halfway along is obvious at print size.
• Background mush — an area that reads as "foliage" at 400px can read as smeared green at 36 inches.

FLUX 1.1 Pro and Gemini 3 Pro Image hold detail best under enlargement. If a piece is destined to be printed large, it is worth the extra credits to render the final on one of them.

Think about where the light falls

A print hangs on a wall in ambient light, not on a backlit screen. Very dark images lose their shadow detail on paper — what reads as moody depth on a monitor reads as a black patch in a hallway. Very bright, low-contrast images go flat.

Images with a clear tonal range — genuine darks, genuine lights, and something happening in between — survive the transition best. Prompting for directional light rather than flat light gets you there almost automatically.

Match the finish to the image

Each finish you enable changes how the same file reads:

• Matte — flattens contrast slightly, kind to painterly and watercolour looks, no glare.
• Lustre — a little more punch and saturation. The safe default for photographic work.
• Wrapped canvas — reads as a painting, hides fine detail. Excellent for loose, textural pieces; poor for anything with small text or delicate line work.
• Acrylic — highest contrast and gloss. Superb for saturated, graphic images; unforgiving of noise and artefacts.
• Rolled — shipped unframed for the buyer to frame themselves. Good for large sizes where shipping framed is expensive.
• Matted — a border is part of the product, so allow for it in the composition.

You do not have to offer every finish on every piece. Enabling only the two or three that genuinely suit an image makes the product page easier to buy from.

Test one before you list twenty

The single most useful thing you can do is order one print of your own work at the largest size you intend to sell. It costs the price of one print and it tells you more about your generation settings than any amount of screen inspection. Most people change something after seeing the first one.

A checklist before listing

1. Generated at the right aspect ratio for the sizes I will offer.
2. Margin at the edges for framing or canvas wrap.
3. Inspected at 100% — faces, hands, repeating structures, any text.
4. Tonal range holds up away from a backlit screen.
5. Only the finishes that suit this image are enabled.
6. Title, page URL and description written for a human, not a search engine.

Next: using print size sets to apply your sizes across the catalogue in one action, and how to price AI art prints.