Photo to Text

You took a picture of something so you would not have to type it. Drop the photo in and get the words back, free, with nothing sent anywhere.

Drop a file
Drop an image or PDF to extract its text
JPEG, PNG, WebP, AVIF, HEIC, BMP, GIF, TIFF, PDF · English (more languages coming) · processed in your browser
PDF pages are rendered and read as images, up to 20 pages per run

Photo to Text — Free, Fast & Private

Everyone does this. You photograph the wifi password taped to the back of the router, the medicine box before you throw it out, a recipe in a friend's cookbook, the parking sign you want to argue with later, a business card, a page of the manual, the timetable pinned to a noticeboard. The picture is now sitting in your camera roll being completely useless, because you cannot search it, paste it, or send the words to anyone without typing them out yourself. That is the job here. Pick the photo, drop it in the box above, and the words come back as text you can copy. It is free, there is no account, and the photo does not go anywhere — the reading happens on the same phone or laptop the picture is already on. Behind the scenes it is a well-established open-source recognition engine that gets downloaded into your browser the first time you visit, along with a language file of about 1.9 MB, and then stays cached so it starts instantly afterwards.

The handwriting question, answered first

Let us get the disappointing bit out of the way, because it is the thing most people are secretly hoping for. If your photo is of somebody's handwriting — a note, a card, a page of lecture notes, a recipe written out by hand — this will not do a good job. The engine learned to read from printed and typed characters, the kind that come off a printer or a screen, where every letter looks the same as every other example of that letter. Handwriting is the opposite of that. Your a is not my a, letters run into each other, and the same person writes differently when they are in a hurry. Very neat, separated capitals sometimes come through. Anything joined-up mostly comes back as words that look real and are not.

The frustrating part is that it will not tell you it failed. It always returns something, so a page of handwriting produces a page of confident nonsense rather than an error message. The confidence number underneath the result is your only warning, and if it is low on a handwritten photo, that is the reason. On printed material — a label, a page from a book, a leaflet, a receipt, a sign — it is genuinely good, and that covers the overwhelming majority of things people point a phone at.

Glare is the number one thing that ruins a photo

Glossy paper, laminated signs, plastic-wrapped packaging, phone screens photographed with another phone: all of them bounce light straight back at the lens, and wherever that bright patch lands the letters underneath simply stop existing. Not blurred, gone. You will look at the photo and think it is fine, because your brain quietly ignores the shiny bit, and then wonder why a whole line came back missing.

The fix takes two seconds. Move so the light source is off to one side rather than behind you, or tilt the page slightly so the reflection slides off somewhere that is not the text. Do not use the flash on anything shiny; it is the most reliable way to blow out the middle of a page. If you are photographing a screen, turn the screen brightness down and get closer instead.

Straight on beats at an angle, every time

Photographing a page from the side is the most natural thing in the world, because you are standing next to it and leaning over. It also does two bad things at once. The page turns into a wedge shape, so lines of text that should be parallel converge towards the far edge, and the recogniser is working out where each line of text runs. Worse, only part of the page is in focus, because the near edge and the far edge are at different distances from the lens and the camera can only choose one. The far half of the photo ends up soft.

Get directly above it. Put the page on a table rather than holding it, stand over the middle, and keep the phone flat and parallel to the paper, so the page fills the frame as a proper rectangle. A slight tilt does no harm at all, and there is no need to be precious about it. But shooting from thirty degrees off, hand-held, is the difference between clean text and a mess, and it costs nothing to take one step to your left.

Books curve, and curved text is hard

A book will not lie flat unless you make it, and the page bends down into the spine. That curve does the same two things as a bad angle, but concentrated in the inner column: the lines bow instead of running straight, and the paper closest to the spine falls out of focus while the outer margin stays sharp. So you get a photo where the right-hand side reads perfectly and the left-hand side is soup, or the other way round.

Press the book open with your hand near the spine, or weigh it down with something. Shoot one page at a time rather than both, and put the spine along the edge of the frame instead of running through the middle. If it is a paperback that really refuses to open flat, photograph the top half and the bottom half separately from straight above, run each one, and stick the results together. Two clean reads beat one bent one.

Watch your own shadow

Standing over a page to photograph it puts you between the page and the ceiling light, which means your head and shoulders land on the paper as a dark band. Recognition works out where the ink stops and the paper starts, and it does that using brightness. A shadow moves that boundary halfway through the picture, so the shaded part of the page can come back garbled while the lit part is perfect. Same problem, different cause, as glare.

Face a window if there is one, or move to where the light comes at the page from the side. Daylight is easier than a single overhead bulb because it arrives from a wide area rather than one point. And fill the frame while you are at it: if half the photo is your kitchen table, half the detail your camera recorded has been spent on wood grain rather than on letters. Get close enough that the text is the picture. Tap the screen on the words to focus there, wait for it to settle, then shoot.

What you get back, and the small print

You get plain text you can select and copy, plus a percentage saying how sure the engine was. Plain text means exactly that: no bold, no headings, no colours, and no columns, so a photo of a menu or a table arrives as a list of things in reading order rather than a neat grid. Check any numbers by eye before you rely on them, because a wrong word is obvious and a wrong digit is not — a phone number or a price with one character swapped looks completely normal.

Two other things worth knowing. It reads English at the moment; more languages are coming, and until then a photo of a page in another language will still hand you something, which is precisely why it is worth saying out loud. And it takes photos, not PDFs: JPEG, PNG, WebP, AVIF, HEIC, BMP, GIF, and TIFF all work, including the HEIC files an iPhone produces by default. When you are finished with the picture, you can shrink it to email it on, or crop it down to just the bit that mattered.

How it works

  1. Check it is printed, not handwritten: Printed labels, books, signs, and receipts read well. Handwriting does not, and it fails quietly by returning text that looks real, so it is worth checking first.
  2. Take the photo from straight above: Put the page on a table, stand over it, keep the phone parallel to the paper, and let the text fill the frame. Tap the words to focus before you shoot.
  3. Kill the glare and your own shadow: Light from the side, not from behind you, and no flash on anything shiny. A bright patch or a dark band across the page will swallow whole lines.
  4. Drop the photo into the box above: It works on a phone as well as a laptop. The first go downloads the engine and a 1.9 MB English file, then it is cached and starts straight away.
  5. Copy the words, double-check the numbers: Prices, phone numbers, dates, and dosages are where a single wrong character hides. Read those back against the photo before you use them.

Frequently asked questions

Can it read my handwriting?
Realistically, no. It was trained on printed and typed characters, and handwriting varies far too much from person to person and moment to moment. Very neat block capitals sometimes work. Joined-up writing comes back as plausible-looking words that are not what was written, so treat any handwritten photo with suspicion.
Why did part of my photo come back as nonsense?
Almost always glare, shadow, or focus. A shiny patch of light or a dark band from your own shoulders shifts the boundary between ink and paper halfway through the picture, and an angled shot leaves the far half of the page soft. Retake it from directly above with light coming from the side and it usually fixes itself.
Does this work on my phone?
Yes. It is a web page, so it runs in the phone browser you already have, and it reads the HEIC files iPhones save by default as well as ordinary JPEGs. Doing it on the phone makes sense when the words are going into a message; do it on a laptop if they are going into a document.
How do I photograph a book page properly?
Flatten it. Press near the spine or weigh the book down, shoot one page at a time from directly above, and keep the spine at the edge of the frame rather than through the middle. If the book will not open flat, photograph the top and bottom halves separately and join the results.
Is it really free, and is there a catch?
It is free, with no account, no daily limit, and no watermark. The reason there is no catch is that there is no cost on our side to recover: your own phone or computer does the work, so nobody is paying for server time to process your picture.
Does my photo get uploaded?
No. The recognition code is downloaded to your browser and runs there, so the picture is read on the device it is already sitting on and nothing is sent out. Once you have visited once, it even works with the wifi off, which is the easiest way to convince yourself.
Can it read a photo of a menu and keep the layout?
It will read the words but not the layout. Prices and dishes come back in reading order as one run of text rather than lined up in columns, because the output is plain text. You get everything that was on the page; you just have to put it back into shape yourself.
Will it work on a photo of a sign in another language?
Not properly yet. English is the only language file included at the moment, with more coming. A photo of a French or Hindi sign will still produce output rather than an error, which is exactly the trap, so a low confidence figure on foreign text is the engine telling you it was guessing.

All Image Tools

Solutions by use case