Convert HEIC to TXT

You photographed a document with an iPhone and now you want the words. That is an ordinary recognition job. What makes this pair different is that on a Windows machine you frequently cannot even see the picture: HEIC holds HEVC, which is covered by patent pools many vendors declined to license, so Windows asks for a paid codec and several browsers show nothing at all.

So the usual advice — convert the picture first, then read the text — needs two steps and a working decoder for the first of them. Both of those live in this browser tab, which is what sidesteps the codec question: converting HEIC to PNG here decodes the HEVC frame in WebAssembly, and the recognition then runs on the PNG.

The recognition engine is handed the file exactly as you dropped it, with nothing decoding it on the way in — deliberately, since pushing an image through a canvas first would flatten transparency to black. That leaves only the readers compiled into the engine itself, and there are six: BMP, JPEG, PNG, PBM, WebP and non-animated GIF. HEIC is not among them, and there is no HEIF reader anywhere in the recognition core this site serves.

So convert the HEIC to PNG on this site, then read the text out of that PNG. The decoding step is a different piece of software from the recogniser — the same one behind HEIC to PNG and HEIC to JPG — and it does handle HEVC stills. Pick PNG rather than JPG for the file in the middle: the photograph has already been through one lossy pass on the phone, and a second one softens exactly the thin strokes the recogniser has to match. Delete the PNG once the text is out.

It would be dishonest not to say this. If the photograph is on the iPhone and you want a paragraph out of it, holding your finger on the text and copying it is faster than anything on a website, and it is built into the phone.

This page is for the other situations: the pictures are already on a computer, there are forty of them, you want the whole page as a file rather than a selection, or you are working on a machine that is not yours. Those are common enough to be worth a page, and Live Text does not help with any of them.

The resolution helps: twelve megapixels or more means plenty of pixels per character even when the page fills only part of the frame. The lens and the processing help too — a modern phone photograph of a page is sharper and more evenly exposed than most flatbed scans from fifteen years ago.

What still governs the result is how the picture was taken. Hold the page flat, shoot straight down rather than from a chair, and get the light from the side so your own shadow is not across the text. Those three things matter more than any setting on this page.

Recognition matches shapes to letters. It does not understand the document, and it produces fluent, complete-looking text whether or not it read every character correctly.

That is the important caution, and it applies most to the things people photograph on a phone: amounts on a receipt, a reference number on a letter, a dosage on a label, a date on a form. Read the output against the original, and read the numbers twice — a wrong digit looks exactly like a right one.

The recogniser works within one language’s letterforms and vocabulary. Given the wrong one it does not report an error; it returns the best match available in a set that does not contain the correct answer, which reads as confident nonsense.

This bites hardest on exactly the documents people photograph while travelling — a menu, a form, a notice, a ticket — where the language is not the one the machine defaults to. If a result looks strange in a way you cannot account for, check this first.

The output is plain text with no layout: no columns, no table structure, no positions on the page. A letter or a receipt comes back in a readable order. A two-column page can interleave, because the recogniser reads across lines that look like lines.

Forms are the hardest case — labels and values end up adjacent in ways that are hard to untangle, and a table of figures becomes a run of numbers with no columns. For those, photographing one section at a time gives a result you can actually use.

Both steps happen inside the browser tab on your own processor: the HEIC is decoded there to make the PNG, and the text is read there too. No upload, no server copy, no account, no daily allowance — and a whole folder of photographs converts as fast as your machine can read them.

For this particular pair that is the substantive point rather than a feature. A photograph of a prescription, a payslip or a signed letter is exactly the file you would not want sitting on somebody else’s server, and the simplest guarantee is that it never went there.