What resolution do AI images come out at, and does upscaling help?

Current models generate around one to two megapixels natively, which is fine on screen and short for print. Upscaling adds pixels and can invent plausible detail, but it also enlarges every existing flaw — fix the image first, then enlarge.

Native output from current text-to-image models sits in the region of a megapixel or two — a comfortable size for a web page, a social post or a preview, and noticeably short of what a printer wants. That gap is where most confusion about "quality" comes from: the image looks perfect at screen size and falls apart the moment it is enlarged, which people read as the model being bad rather than as an ordinary resolution problem.

What you actually need

UseRule of thumbComment
Web, social, in-appNative output is enoughCompression will cost you more quality than resolution does
Small print up to A5~150–300 dpi at final sizeNative output often just clears this
A3 poster and upUpscale requiredTwo to four times native, at minimum
Large format, viewed closeUpscale plus manual retouchingInvented detail becomes visible at this size

What upscaling really does

A modern upscaler is not stretching pixels. It is a model that hallucinates plausible detail into the gaps: pores where there was a smooth cheek, fibres where there was a flat coat. That is why the result can look sharper than the original in a way that classical resizing never achieves — and also why it is an unreliable narrator. Fine text becomes confident nonsense, and an ambiguous shape becomes a definite wrong thing.

  • Anything you would not want invented — logos, faces of real people, small text — should be fixed before the upscale, not after.
  • Artifacts that are borderline at native size become unmissable at four times the pixels. Judge the source at 100 percent, not fit-to-screen.
  • Upscaling a heavily compressed JPEG upscales the compression too. Always start from the largest, least-processed file you have.

Aspect ratio is not resolution

Choosing a portrait or landscape frame changes the shape of the output, and usually keeps the total pixel count in a similar band. A tall image is not a higher-resolution image — it is the same budget of pixels arranged differently. If you need both a specific shape and a large size, plan on generating in the right shape and upscaling afterwards, rather than expecting the frame choice to buy you detail.

Judge sharpness at 100 percent zoom on the actual file. Screens scale images down by default, and downscaling hides exactly the class of soft, mushy detail that ruins a print.

More on this

Other questions

← all questions