Image to text converter

Reembun. (2026, July 31). Image to text converter. https://reembun.com/image-to-text

What it reads

Input	How it is handled
JPG, JFIF, PNG, GIF, WebP, BMP	Decoded by the browser, then read
HEIC and HEIF from an iPhone	Decoded here, because no browser except Safari will
PDF with real text	Text taken directly, with no recognition step
PDF that is a scan	Every page rendered at 200 dpi, then read
Screenshot on the clipboard	Paste it with Ctrl+V anywhere on this page
A photo on your phone	The camera button opens the rear camera

Multi-page PDFs are expanded automatically, page by page, and the results are joined in reading order.

Two output modes, and when to use each

Simple returns the engine’s own output. Line breaks land where the page broke them, which is what you want when you are quoting a paragraph or copying a serial number.

Formatted reads the geometry rather than only the characters. Every line carries a position and a height, and that is enough to recover the structure:

• A line set noticeably larger than the body text, sitting on its own and not ending in a full stop, becomes a heading.
• Lines that begin with a bullet or a number become a list, and consecutive items are folded into one.
• Lines that split into the same number of columns at the same positions become a table.
• Wrapped lines are rejoined into paragraphs, and a word broken across a line break is put back together.

You can then take that as a Word file, plain text, Markdown or HTML. The Word download is the one to use when the text has to be edited or sent on, because the headings, lists and tables arrive as real Word headings, lists and tables rather than as characters that look like them. Markdown suits a note-taking app or a CMS, since the structure survives the paste.

Why the photo matters more than the settings

Recognition failure is almost always an image problem. In rough order of how much damage each does:

Angle. A page photographed from 30 degrees off makes every character a different shape than the one the engine learned. Straighten up before you reach for any setting.

Shadow. A hand or a phone casting a line of shadow across the page splits it into two exposure zones, and the darker one usually reads as nothing at all.

Distance. Text needs to be roughly 30 pixels tall to be read reliably. Standing back and cropping later throws that away, because the pixels were never captured.

Motion. A slightly blurred photo looks fine at a glance on a phone screen and is unreadable to an engine that works letter by letter.

The confidence figure under the result is the engine’s own average across every word it read. Below 70 it is telling you plainly that it is guessing, and the fix is another photograph rather than another setting.

Page layout is a real setting, not a preset

If the words come back scrambled, the layout selector is the thing to change. Automatic detection tries to find columns and reading order, which is right for a book page and wrong for a receipt with a single narrow column of prices. Scattered text turns off layout analysis entirely, which is what a photo of a shop sign or a screenshot of a dialog box actually needs.

The privacy difference is structural

Every other free OCR site works the same way: your image is uploaded, processed on a server, and you are asked to trust a sentence about deletion. That sentence may well be true. It is unverifiable either way, and the image was still transmitted, still sat on a disk, and still passed through whatever logging that company runs.

Here the engine comes to the image. It is about 4 MB and downloads once, then your browser keeps it, so the second image you read starts instantly and works with no connection at all. That is not a marketing position. It is the only arrangement under which a payslip, a medical letter or a signed contract can go into an online OCR tool without leaving the building.