zerouploads

Image to Text (OCR)

Read the text in a photo, screenshot or scan. Words the reader is unsure of are marked, not quietly guessed.

  • Unlimited
  • No signup
  • Private
  • Works offline

Read the text in an image

No image yetdrop, paste or choose an image to start

Choose an image file

Sourcedrag to move · scroll or pinch to zoom

no image yet

Recognised text

The text appears here, in the order it was read.

——

Reading settings
The text appears once an image is read

How it works

  1. 01

    Add an image

    Take a photo on your phone, or drop, paste or choose a file. The reader loads once, then works offline.

  2. 02

    Check the regions

    Each block is outlined on your photo. Scroll or pinch to zoom in and check a word against the page.

  3. 03

    Fix, then take it

    Marked words show where to look. Edit them, then copy the text or save it as .txt or Word.

What affects accuracy

Size matters more than anything else. A letter needs about 20 pixels of height to be read well. A 300 DPI scan of a printed page gives you that. A photo of a page usually clears it too. A screenshot of small type often does not, and neither does a crop from a web page.

Straight, evenly lit pages read better than angled ones. If half the page sits in shadow, the words in the shade come out worst. Retake the photo from above with the page flat, and most of the errors go away.

Handwriting is the one thing this cannot do. The model learned printed type, so joined-up writing comes out as noise. Printed receipts and forms work well, with one catch. The letter O and the digit 0 look the same in many fonts. So do the letter I and the digit 1. Those words score low, so the page marks them instead of showing them as read.

The figure in the corner is the mean score for every word found. Read it as a rough guide. A low score across a whole page nearly always means the image is too small or too dark. Fix that first, because a sharper image beats every setting on this page.

Languages
English only. Each language needs its own model, and the English one is about 2 MB.
Model
Tesseract, built to WebAssembly. Served from this site, and kept in your browser after the first run.
Layout
Auto-detect columns finds columns and blocks for you. Page layout can also be One column, One block of text or Scattered words.
Confidence
A score for every word. The page marks any word under 70%, so you check it instead of trusting it.
Editing
The text box is yours to correct. What you copy or save is what you see in it.
Output
Plain text, or a Word file. The Word file is monospaced, so columns stay lined up.

Frequently asked questions

Is my image uploaded to be read?
No. Your browser downloads the reader once and runs it on your own machine. Receipts, contracts and ID papers never leave it. You can turn off your network after the first load and the tool still works.
Can I take a photo with my phone?
Yes. On a phone or tablet there is a Take a photo button, which opens your own camera app. The photo goes straight into the reader and is never sent anywhere. Hold the paper flat and fill the frame with it.
My photo is a bit crooked. Does that matter?
A few degrees is fine. Past that, turn on Straighten first, and the reader squares the page up before it reads. The outlines go away while it is on. The reader measures them on the straightened page, so they would sit at an angle to your photo.
Can I fix a word the reader got wrong?
Yes. Press Edit and the text becomes a box you can type in. Copy and download both take your version. Undo edits puts the reader's own text back, marks and all.
Can I save it as a Word file?
Yes. Set Output to Word (.docx), and you get a file that Word, Google Docs and LibreOffice all open. It uses a monospaced font so a receipt's columns stay lined up.
Why is some of the text wrong?
Small, blurred or dim letters are guesses, and a guess is sometimes wrong. The words it was least sure of are marked for you. A bigger, sharper image of the same page fixes most of them.
Does it read handwriting?
No. The model was trained on printed type. Neat block capitals sometimes come through, but joined-up writing does not. Treat handwriting as out of scope here.
What size should the image be?
Aim for text about 20 pixels tall. On a printed page that is a 300 DPI scan, or a phone photo taken close in. Very large images are fine up to 40 megapixels, but they take longer.
Can it read two columns?
Yes. Auto-detect columns finds them, and the text comes out one column after the other. If a page still reads as a jumble, set Page layout to One column and try again.
Can I read a PDF?
Not here. Save the page you need as a PNG or JPG first, then drop that in. If the PDF has real text in it, PDF to Text reads it straight away.
Which languages does it read?
English only. Every other language needs its own model, and shipping all of them would make the download much bigger for everyone.
Why is the first read slower?
The first one fetches the reader and the English model, about 5 MB in total. Your browser keeps both, so the next image starts straight away.
Is there a limit on how many I can do?
No. There is no daily count, no sign-up and no watermark. One image at a time, because the whole page is held in memory while it is read.

More image tools

Text DiffCompare two versions of a text and see exactly what changed, word by word or line by line — in your browser.AVIF to JPGConvert AVIF images to JPG in your browser. AVIF packs tighter than JPG, so expect a bigger file, and clear parts take the colour you pick.Compress Image to KBShrink a photo to the kB limit a form asks for. Quality drops first, then the picture if it must, and nothing is uploaded.GIF to PNGConvert GIF images to PNG in your browser. An animated GIF gives you its first frame, and clear pixels stay clear.JPG to PNGConvert JPG images to PNG in your browser. PNG keeps every pixel, so expect a bigger file. The queue shows each new size before you save.JPG to WebPConvert JPG photos to WebP in your browser. Quality 82 looks the same to most eyes, and the queue shows each new size before you save.
Browse all image tools