Handwriting OCR, on your device
Handwriting to text
Turn a photo of handwritten notes into text you can edit and search. The page is split into lines and read by a handwriting model running inside your browser — free, private, and offline after the first load.
Drop a photo of your handwritten page
JPG or PNG. The page is split into lines and read on your device — nothing is uploaded.
The model downloads once and is then cached by your browser for offline use.
Built for handwriting
A transformer trained on handwritten lines, not a print-OCR engine forced to guess at cursive.
Nothing uploaded
Diaries, notes and forms are read by your own processor. No server ever sees the page.
Works offline
The model is cached by your browser after the first load, so later pages need no connection.
How it works
- 1
Photograph the page flat
Even light, no shadow across the paper, and the page filling the frame. Handwriting recognition is far more sensitive to a skewed photo than printed text is.
- 2
Pick a model
Start with the fast model. If your writing is cursive or cramped, switch to the accurate one — it is a larger download but noticeably better.
- 3
Let it read line by line
The page is split into individual lines of writing on your device, then each line is recognised in turn, appearing as it completes.
- 4
Fix and keep
The text box is fully editable. Copy it, or save it as a .txt file.
Common questions
Why does normal OCR fail on handwriting?
Classic OCR engines like Tesseract match glyph shapes against printed typefaces, and handwriting has no fixed shapes — letters join, slant and change with every writer. This page uses a transformer model trained specifically on handwritten lines, which reads a whole line in context rather than letter by letter, so it copes with joined-up writing.
Is my page uploaded anywhere?
No. The model is downloaded into your browser once and then runs on your own processor — with WebGPU when your device offers it, otherwise WebAssembly. Your notes, journal or medical form never leave the device, and after the first load the page works with the network switched off.
How accurate should I expect it to be?
Neat, well-spaced handwriting on unlined or lightly lined paper typically comes back highly readable with a handful of corrections. Cursive, cramped or faded writing is harder — treat the output as a strong first draft that saves you the typing, not as a perfect transcription.
Which languages does it handle?
The models here are trained on English handwriting. Printed text in other languages is better served by the OCR studio, which supports thirteen languages.
Why is the first run slow?
The recognition model has to download the first time, which is a one-off cost of a few hundred megabytes depending on the model you choose. Your browser caches it afterwards, so every later page is fast and works offline.