AI · 6 min read

AI business card design: what actually happens

“AI business card maker” covers two completely different things. One generates a picture of a card. The other generates a card. The difference does not show up on screen — it shows up at the press.

The problem it is actually solving

Almost nobody is stuck because they cannot operate a design tool. They are stuck on the blank canvas: what colour, what layout, how big the name, what goes in the empty half. That is a taste problem, and it is the one thing a model is genuinely good at — it has seen more business cards than any of us ever will.

So the useful job for an AI here is narrow and specific: turn a sentence about your trade into a starting point that is already close, then get out of the way.

A picture of a card is not a card

This is the part worth understanding before you pick any tool. An image generator produces one flat grid of pixels. Your name is not text in that file — it is a smear of coloured dots that happens to look like text at screen size.

Three things follow, and all three are bad. You cannot fix a typo without regenerating the whole card. The type prints soft, because it is being resampled from a photo rather than drawn from an outline. And models are famously unreliable at spelling inside images, so the name on the card is not guaranteed to be your name.

The other approach generates the background as an image, then works out the layout separately and returns it as structured data: this text, this font, this size, at these coordinates, in this colour. That is what opens in the editor. Your name is text the whole way through, so it is spelled correctly because you typed it, it is editable because it was never flattened, and it is rendered from the actual font outline when the print file is built.

Why the layout step is separate

Once a background exists, the interesting question is where the type can go without disappearing. A generated background is not flat — it has a bright corner, a dark gradient, a photographic area that is busy and a clean area that is not. Black text over the wrong third of it is unreadable.

So the background gets looked at before the type is placed: find the calm regions, measure how light or dark they are, and pick text colours with enough contrast to survive. That is a separate decision from “make a nice background”, and doing it as one step is why so many AI-generated cards have a beautiful image with a name you cannot read across it.

What it gets right, and what you will still fix

Being straight about this is more useful than a sales pitch:

• Reliably good: colour palettes that go together, an overall mood that matches a trade, a sensible split between artwork and clear space, and getting you off the blank page in seconds.
• Usually needs a nudge: the exact size of your name relative to everything else, and how much air sits around the contact block. Models tend to run type slightly large.
• Always yours to decide: which single call to action belongs on the card, what goes on the back, and whether the whole thing actually sounds like you.

Treat the first result as a rough draft that arrived in four seconds instead of an afternoon. The value is the head start, not the finality.

How to describe a card and get something good

Vague inputs produce generic output, and “make it professional” is the vaguest input there is. Three things sharpen it immediately:

Say what you do, not what industry you are in

“Mobile dog groomer, home visits” produces a very different card from “pet services”. Specifics carry visual information; categories do not.

Name the feeling, and name its opposite

“Warm and hand-made, not corporate” steers far harder than “modern”, because it rules something out. Almost every card ever made is “modern”.

Give it one colour to anchor to

If you already have a logo or a van livery, say the colour. It is the cheapest way to make the card look like it belongs to the rest of your business.

Then generate two or three and pick, rather than trying to talk one result into perfection. Comparing options is much easier than describing what is wrong.

The bit that decides whether it prints

None of the above matters if the file is wrong at the press, and this is where a card built as real objects earns its keep. The canvas is 3.625″ × 2.125″ at 300 dpi, which is bigger than the finished card on purpose — the background runs past the cut so there is no white sliver when the blade drifts, and your type stays inside a safe area well clear of it. Colour is converted for print rather than trusted from the screen, which is why a glowing screen blue never comes back quite the same.

You do not have to think about any of it, but you should be able to see it. The editor draws those lines while you work, and the proof is the real artwork at real size, not a stock photo of a card with your logo pasted on.