Utilore

Image to Text

Drop in a picture, pick the language of the writing, and press the button. A scan or a screenshot usually comes out word for word; a photo taken by hand comes out almost right and wants a read-through.

or drop them here

A picture with text — a scan, a screenshot or a photo. JPG, PNG, WebP, HEIC, up to 50 MB

The recognised text is given exactly as it was read — nothing is auto-corrected, because a plausible correction to a number is a mistake you can no longer catch. Layout is not preserved either: columns, tables and headers come out as ordinary lines. The engine and the language file are served from this site, not from someone else’s, and the picture itself is never uploaded — a document photographed on a desk is precisely the kind of file that should not travel.

How to use

  1. Drop a picture onto the area below, or press «Select file» — JPG, PNG, WebP and HEIC from a phone.
  2. Pick the language of the text. The wrong language is the most common reason nothing is found.
  3. Press «Read the text», then copy it or save it as a .txt file.

Good to know

How well it actually reads, measured rather than promised

On a clean scan or a screenshot the text comes out character for character. On a photo taken by hand — tilted a little, lit unevenly, compressed by the camera — between one and ten characters in a hundred come out wrong, and Russian fares noticeably worse than English on the same frame. The mistakes are the predictable kind: a «g» read as «§», a hyphen becoming a dash, two words running together. That is why the page tells you which of the two cases you are in instead of handing back the text as though it were typed.

Nothing is corrected, and that is on purpose

It would be easy to notice that «37-70» on a receipt is almost certainly «37.70» and quietly fix it. That is exactly what this page does not do. A plausible correction to a number is a mistake you can no longer catch by reading — and numbers are where it costs most. What you get is what was read. Layout is not kept either: columns, tables and running heads come back as ordinary lines, because a picture holds no information about which of them was which.

Frequently asked questions

Is my picture uploaded anywhere?

No, and neither is anything else. The recognition engine and the language file are served from this site rather than a third-party CDN — otherwise the mere fact of your visit would be known to someone else. The picture itself never leaves the browser. A document photographed on a desk is precisely the kind of file that should not travel.

Can it read handwriting?

Not usefully. The engine is trained on printed and typed text; on cursive it produces something that looks like words and is mostly wrong. Block capitals written very evenly sometimes come through. Anything else is a coin toss, and a coin toss dressed as a result is worse than a refusal.

Why does the first run download three megabytes?

Because the recognition runs on your device, and that takes an engine and a model of the language. Together they are about three megabytes, said before the button rather than after. The browser keeps both, so the second picture starts immediately — and works with the connection off.

I have a scanned PDF, not a picture.

Turn it into pictures first with the «PDF to JPG» page and bring those here. If the PDF has a real text layer — most documents made on a computer do — then «PDF to Word» will give you the text without any recognition at all, and without its mistakes.

Related tools