Image to text
Read the words out of a screenshot, scan or photo — on your own device.
- Runs on your device — the file is never sent anywhere
- No sign-up, no watermarks, no daily quotas
- Free, with no limit on how often you use it
How it works
Optical character recognition is the tool people reach for with a screenshot they cannot copy from, a scanned page they need as text, or a photo of a document. It is also the one where uploading is hardest to justify — the images are usually documents, receipts or correspondence.
It runs on your device, which is unusual for OCR
Almost every free OCR service uploads the image and does the work on a server. This runs Tesseract compiled to WebAssembly in the page. The trade is a one-off download of about 11 MB of recognition data on first use, which your browser then caches, and the recognition itself using your own processor.
What it is good at, and what it is not
Screenshots, clean scans and printed documents photographed straight-on give very good results. Text at an angle, in poor light, over a busy background, or in an unusual display font gives poor results. Handwriting is not supported in any practical sense. It reads English.
Layout settings matter more than people expect
"A page of text" suits documents with paragraphs. "A single block" is better for a screenshot of one passage. "A single line" is for a caption or a serial number. "Scattered words" suits signs and labels with text in several unrelated places. Picking the wrong one is a common reason for poor output.
Getting a better result
Crop to just the text before running it. Rotate the image so lines are horizontal. Higher resolution helps up to a point — around 300 dpi equivalent is ideal, and beyond that adds time without accuracy.
Image to text — questions
Is my image uploaded?
No. The recognition engine runs in your browser. The image never leaves your device; only the engine itself is downloaded, from a public open-source CDN.
Why is it slow the first time?
About 11 MB of recognition data is downloaded once and then cached by your browser. Later runs skip that.
Does it read handwriting?
No. Handwriting recognition is a different problem and this does not attempt it.
Which languages does it read?
English. Other languages would each need their own recognition data downloaded.