Can ChatGPT or Claude make a poster or flyer?

It can make an image of a poster. Ask for one and you get something plausible back in under a minute, and for a mood board or an idea that is useful. The gap between that and a poster you can put in a window is the text.

Text rendering in image models improved a great deal through 2025. Short headlines in English now come out right more often than not. It is the rest that fails: a phone number with one digit changed, a price rendered as $l2, a second line of body copy that dissolves into letter shaped marks, an apostrophe in the wrong place.

The tell is that you cannot fix it. There is no text layer, so the only repair is another generation, which changes everything else too.

Claude does not generate images. What it can do, and this is more useful than it sounds, is write the code for one. Ask for an SVG or an HTML page and you get a document where the text is real text, positioned deliberately.

That is closer to a real poster than a generated image, because you can change a date by editing a line. The catch is that you now have a file you have to render, and the visual judgement is uneven. It has no measured sense of whether your headline actually fits the box, so long words overflow, and gaps collapse when text wraps in ways the model did not anticipate.

Because an image model is painting, not typesetting. It has no font file, no baseline, no letter spacing, and no concept that a particular string must survive intact. It has learned what text tends to look like, so it produces something with the texture of writing.

That explains the pattern people notice. Common short words come out fine because the model has seen them a million times. Prices, phone numbers, dates, and proper nouns come out wrong because they are arbitrary strings, and arbitrary is exactly what a texture model cannot do.

It also explains the second language problem. Ask for a Spanish version and you frequently get Spanish shaped marks: an accent in the wrong place, a word that is nearly right. If you cannot read the language yourself, you will not catch it, and the person who can read it will.

Everything, which is the honest answer. A generated image is a single flat picture. Ask for the same poster with Friday instead of Thursday and you get a new poster: different composition, different colours, different illustration. The one you liked is gone.

This matters more than the initial quality, because real use is mostly changes. The price went up. The date moved. You want it in Spanish for the other window. Every one of those is cheap in a document and a fresh roll of the dice in an image.

Three cases, and they are real. Working out what to say. Describe your event to any of these models and ask for ten headline options. That is a language job and they are good at it.

Mood boards. If you are trying to explain a look to a designer, generated images are a fast and cheap way to point at what you mean.

Backgrounds and textures. An abstract background with no text in it has none of the failure modes above. If you need a photographic or painterly image to sit behind real type, that is a good use.

Separate the layout from the art. The words are rendered as real text in a real typeface. The illustration is vector art placed in a slot. The result is a document rather than a picture, so a change touches only what you changed.

That is how Dovey works, and it runs over text message rather than in an app. You text what you need, the poster comes back in the same thread in well under a minute, and "make it Spanish" translates the words while the layout stays where it was. It is free for your first 3 designs or edits, with no card.

The honest limits are the same ones this whole article has been about, pointed the other way. You get a PNG, not a print-ready PDF. The pictures are Dovey's own, painted in a printed, hand-drawn or photographic style, or your photos if you text them. And if you want to place every element yourself, you want an editor, not a text thread.