Image to text (OCR)
Read the text out of a photo or screenshot.
Drop your file here
Processed securely and deleted within the hour.
Reading text out of a picture
Photograph a page, a receipt, a whiteboard, a slide or an error message, and this reads the characters out of it and gives you the text. It is the fastest route from something printed to something you can paste.
Getting a good result
Recognition quality depends far more on the photograph than on the software, and the difference between a poor capture and a good one is enormous.
- Shoot straight on. A page photographed at an angle has letters that lean and stretch. Directly overhead is worth more than any setting.
- Even light. A shadow across the page or a bright reflection off glossy paper defeats recognition locally, so part of the text comes back and part does not.
- Fill the frame. Text occupying a small part of a photograph has few pixels per character. Get closer rather than cropping afterwards.
- Hold still. Motion blur is the single most common cause of a poor result.
Keeping the layout
By default, lines are kept as lines, in reading order. Turning that off joins everything into one flowing block, which is what you want when the source is prose that happened to be broken across lines and you intend to reflow it anyway.
Columns, tables and receipts are inherently harder — recognition returns positioned fragments, and turning positions back into a table is a separate problem. Expect a sensible line-by-line reading rather than a reconstructed layout.
Confidence is reported
Every result comes with an average confidence figure, and low confidence is flagged. That matters because OCR fails plausibly: it does not return gibberish, it returns a word that looks similar. Check figures and names against the image before relying on them — 0 and O, 1 and l, 5 and S are the usual confusions.
For a whole document
If your source is a multi-page scanned PDF, OCR PDF handles it in one pass and gives you back a searchable document rather than loose text.
Common questions
What image quality do I need?
Straight-on, evenly lit, filling the frame, and sharp. That matters far more than any setting here.
Does it read handwriting?
Poorly. The models are trained on printed text.
Can it keep table layout?
Not reliably. Recognition returns positioned lines; rebuilding a table from positions is a separate problem.
Which languages work?
Latin scripts and Chinese are handled well by the bundled models. Other scripts vary.
Why is one word wrong?
OCR fails plausibly rather than obviously. Check digits and names — 0/O, 1/l and 5/S are the usual confusions.
Is my image uploaded?
Yes, this one runs on our server. It is deleted as soon as the result is sent.