Image to Text
NewScan a page or your handwriting into editable text, in your browser.
Runs entirely in your browser
Loading tool...
How to use
- Open the camera and point it at the page, or drop in a photo you already have.
- Capture as many pages as you need; they are read in the order you add them.
- Choose whether the page is printed text or handwriting, and set the language.
- Select Extract text, then copy or download the result.
Features
- Reads printed text in English and Vietnamese, and English handwriting.
- Runs entirely on your device — no photograph is ever uploaded.
- Straightens a tilted photo and finds the lines before reading them.
- Scan several pages in one go and get a single block of text.
- Works offline once the recogniser has loaded.
Frequently asked questions
- Are my photos uploaded?
- No. Both recognisers are WebAssembly and run inside the tab, and the tool has no upload endpoint to send an image to. That is the point of building it this way: people scan prescriptions, letters and notebooks, and none of that should be sitting on someone else's server.
- Why does it not read Vietnamese handwriting?
- The handwriting model was trained on an English corpus, and no Vietnamese equivalent is published in a form a browser can run. Fed Vietnamese it produces English-shaped guesses with the tone marks missing, so the tool says so rather than offering it. Printed Vietnamese is a different matter and is read well.
- How accurate is handwriting recognition?
- Far less accurate than printed text, and that is a property of the problem rather than of this tool. Neat, separated writing on unlined paper in good light reads well; joined or hurried writing often does not. Read the result against the page before you rely on it.
- Why is the first scan slow?
- The recogniser has to load before it can read anything — a few megabytes for printed text, considerably more for handwriting. Your browser caches it, so a second scan starts immediately, and the tool then works with no network at all.
- How do I get a better result?
- Light the page evenly and avoid your own shadow, fill the frame with the page, and hold the camera parallel to it. A small tilt is corrected automatically. If the text comes back as nonsense, the photograph is almost always the reason rather than the settings.
- The camera will not open. What now?
- The tool reports which of the three reasons it was: you or your browser refused permission, the device has no camera, or another application is already using it. Permission can be restored from the padlock in the address bar. Uploading a photo works regardless and is often better on a laptop anyway.
Related tools
Turn an image into a data URI you can paste into HTML or CSS — and back again.
Fix text that has been through a PDF, an email client or a word processor — one decision per switch.
Count words, sentences and reading time as you type.
Translate an English PDF into Vietnamese inside the original file, with the layout, the images and the page count left alone.
Flesch, Fog, Kincaid and SMOG on your text — plus the specific sentences making it hard to read.