You walk into a print shop with a file that looked expensive on your phone. The person behind the counter opens it, zooms once, and sends you home. Not because they hate AI. Because the file you brought is a screen picture pretending to be a print job.
That happens a lot now. Midjourney, ChatGPT image tools, and the rest are very good at making something you want to hang on a wall. They are not making the file a press needs. Those are different jobs, and the shop is paid to notice.
Print shops talk about dots per inch because a press lays down ink on paper of a known size. Midjourney's own image size notes do the same math in public: a 2048 pixel square at 300 dpi is about 6.8 inches on a side. OpenAI's image models sit in the same neighborhood. DALL-E still ships common sizes around 1024 pixels, or 1792 by 1024 for a wide frame.
Color is the other half. Screens use RGB. Offset and most digital presses use CMYK. The overlap is smaller than it looks. Bright RGB colors sit outside the CMYK gamut and have to be mapped to something ink can mix.
The missing page furniture has names. Bleed is the extra image past the trim so a slightly off cut does not leave a white sliver. Many shops want a PDF/X or at least a print PDF, not a chat export. Canva documents the split in the open: PDF Standard is 96 dpi for screens, PDF Print is 300 dpi, with bleed and crop marks. Their bleed help adds about 3 mm on each side.
They are counting pixels, not vibes
A typical generator still starts around 1024 pixels on a side. Some paid modes go to 2048. That number is the whole story. Print quality is pixels per inch at the size of the paper, not a 300 stamped in the file info.
Do the division. 1024 divided by 300 is about 3.4 inches, a little under 9 centimeters. That is a large postcard if you are lucky. An A4 flyer wants roughly 2480 by 3508 pixels at 300 dots per inch. Stretch your square masterpiece across that sheet and you are closer to 120 dpi. Fine on Instagram. Soft in someone's hand.
Changing the dpi number in an editor without adding pixels does nothing useful. You made the document claim it is smaller. The shop still sees 1024 pixels.
The color you saw is not ink
Screens mix light: red, green, blue. Presses mix ink: cyan, magenta, yellow, black. The RGB box is bigger. Neon teal, electric purple, and the glowing sunset the generator loves often sit outside what CMYK can print. The conversion does not ask permission. The color just goes dull, or it shifts toward a neighbor.
If your brand lives in one of those impossible blues, the shop is not being difficult. The ink cannot do that trick.
There is no flyer in that file
A generated picture is a rectangle of pixels. A print job is a page with a size in millimeters, extra image past the trim so the cut can miss by a hair, a safe zone for type, and usually a PDF the shop software understands.
AI does not add 3 mm of extra picture around an A5. It does not keep your phone number 5 mm inside the cut. It does not export a print PDF. You download a PNG from a chat window and call it a poster.
The letters are part of the picture
Ask a generator for a poster with a headline and you will get melted type, swapped letters, and a phone number that almost exists. That is not a font. You cannot fix the spacing. At print size those mistakes are the first thing a human reads.
Same problem with logos. A mark that has to live on a van and a pen needs to be a vector: lines and type you can scale. A 1024 pixel "logo" is a photograph of a logo. The shop that asks for an SVG or a PDF of the mark is asking for something you can still edit at any size.
The weird bits get huge
Hands with extra knuckles, repeating tiles in a sky, plastic skin. On a phone you scroll past. On a 50 centimeter poster they are the design. Shops that have printed a few of these start rejecting on sight, because reprints come out of their pocket or yours.
How to get it printed anyway
Stop treating the generated file as the job. Treat it as a photo you will place inside a real layout.
Build the page in Canva (or InDesign, or Affinity Publisher). Those tools exist for type, margins, and a finished size in millimeters. The generator does not. Open a custom canvas at the exact print size, turn on bleed, and put every word in as real text. Your logo goes in as a vector or a sharp PNG. The AI picture is a background or a panel, not the whole flyer.
Before you drop that picture in, make the pixel math work. Inspect it at 100 percent zoom. If the face or the product edge is already soup, do not print it. Then upscale so a 1024 image can cover more paper. Start at 2x. Check faces, edges, and any accidental letters again at 100 percent. Upscaling invents plausible detail. It will not turn melted type into a real sentence, and it will not invent a logo that was never drawn as one.
Place the enlarged file in the layout and do not stretch it past what the pixels allow. Canva will warn you when an image is too small for the box. Believe the warning. Leave a quiet margin so type does not sit on the trim. Export PDF Print with crop marks and bleed. If Canva offers a CMYK profile, use it. If the shop prefers to convert, ask them and write that down. Do not send the PNG from the chat.
That split is the whole workaround. The generator makes the scene. The layout tool makes the page. The shop gets a file that looks like work.
When it still will not work
A 1024 image blown up for a shop window is a rumor of a picture. Viewing distance helps a big banner more than a flyer in the hand, but it does not excuse a file that was never meant to be that wide. A generated logo for embroidery or vinyl is the wrong kind of file. Tiny type baked into the art cannot be rescued. If the color is the product, pay for a printed proof before you order 500.
The generator makes the scene. Canva (or any layout tool) makes the page. Upscale the picture until the pixel math works, keep the type as type, export a print PDF, and take that to the counter. The shop was never rejecting the idea. They were rejecting a screenshot of one.