Copy the text out of a screenshot

Why screenshots read better than photographs

A screenshot is the easiest thing text recognition ever gets. The letters were drawn by the same machine that saved the file, so they are perfectly sharp, perfectly level, evenly lit, and rendered in a clean typeface at consistent spacing. None of the problems that dominate reading a photograph — blur, skew, shadows, uneven lighting, paper texture, camera noise — exist at all. Accuracy on a good screenshot is dramatically higher than on a good scan, and far higher again than on a phone photo of a screen.

Resolution is the whole game, and it is the one thing you control. A letter needs roughly sixteen pixels of height to be read reliably, and below about eight there is simply not enough detail to reconstruct — enlarging afterwards cannot add information that was never captured. So a screenshot taken on a high-density display reads beautifully, and the same screenshot shrunk to fit a document reads badly. Capture at full size, crop rather than scale, and never re-screenshot a screenshot.

Format matters more here than for photographs, for the same reason. PNG keeps every edge exactly as drawn; JPEG does not — its compression puts ringing around high-contrast edges, and text on a plain background is nothing but high-contrast edges. A JPEG screenshot of small text can be measurably harder to read than the PNG of the same thing, so if your screenshot tool offers the choice, PNG is correct.

What still defeats it is text that was never meant to be read as text: words over a busy photograph, light grey on white, letters in a decorative or condensed display face, or a dark-mode interface where the contrast is inverted and low. The result is plain text in any case — typefaces, colours, bold, italic and layout are not carried, because what comes out is the words rather than a copy of the interface.

When you want something else

Your operating system may already do this better and instantly. macOS Live Text lets you select text directly in any image in Preview, Photos or Quick Look with no conversion step at all; Windows PowerToys Text Extractor grabs text from anywhere on screen with a keyboard shortcut; Google Lens does the same on Android. For a one-off grab, those are faster than any page.

For screenshots of code, recognition is the wrong approach altogether and it is worth saying so. The characters that matter most — brackets, quotes, pipes, the difference between l, 1 and I, or 0 and O — are exactly the ones most often confused, and a single wrong character in code is a bug rather than a typo. Find the original source if there is any way to; retyping is more reliable than correcting.