Take a photo of a handwritten line, load it below, and turn it into editable text - the AI model runs on your device, so the photo never leaves your browser.
Example - what a result looks like (no download needed to view this):
Transcript the model produced: meet-me at the library tomorrow
Small slips like the stray hyphen above are typical - the transcript is an editable first draft, so review it before you use it.
Handwriting to Text Converter (AI OCR)
This tool turns a photo of handwritten notes into editable text using an AI model that runs entirely on your device. Load the photo, click once, and copy or download the transcript - the image never leaves your browser, because there is no upload and no server.
How it works
The first time you convert something, the page downloads a handwriting-recognition model (about 65 MB - the size is shown before anything starts) and caches it in your browser, so later runs skip the download. The model reads one line of handwriting at a time; when your photo holds several lines, the page slices it into line bands automatically and transcribes them in order, top to bottom.
What it is good at, and what it is not
It is built for real pen-on-paper handwriting in English - meeting notes, to-do lists, a recipe card, a letter. It produces a solid first draft, not a perfect copy: expect the occasional wrong word or stray hyphen, and review the transcript before you rely on it.
- Reads one language (English) and does not detect others.
- Dense, overlapping, or heavily slanted script lowers accuracy.
- Not a printed-text reader - for screenshots, scans of typed documents, or book pages, the image to text OCR tool is the right pick, because it uses a print-trained engine and handles that job better.
Getting a better transcript
- Photograph the page straight-on in even light.
- Keep one line of writing per row and crop away anything that is not text.
- A plain background helps the line slicer find each row.
- If a line comes back as nonsense, crop the photo to just that line and run it again.
Privacy
The photo is processed with WebAssembly or WebGPU inside the page. Nothing you load here is sent anywhere - the only network traffic is the one-time model download from a public model CDN.
Frequently Asked Questions
Is my handwriting photo uploaded anywhere?
No. The recognition model runs inside your browser with WebAssembly or WebGPU, so the photo stays on your device. The only network traffic is a one-time download of the model itself (about 65 MB), which is then cached for later visits.
How accurate is the transcript?
On clear pen-on-paper handwriting it usually gets whole lines right, but treat the result as a first draft: the occasional wrong word, missing punctuation, or stray hyphen is normal. Review and edit the text before you rely on it.
Can it read typed or printed text, like a screenshot or a scanned document?
That is not what this model is trained for, and results on printed text are poor. Use the image to text OCR tool instead - it runs a print-trained engine and handles screenshots, scans, and book pages properly.
Does it work on cursive, and on languages other than English?
It handles joined-up English handwriting, which is its training ground. It reads English only and does not auto-detect other languages. Very dense, overlapping, or heavily slanted writing lowers accuracy.
Why does it read my note line by line?
The model transcribes one line of handwriting per pass. The page automatically slices a multi-line photo into horizontal line bands and reads them top to bottom, then joins the results. If a line is missed, crop the photo to just that line and run it again.
Does it work offline?
After the first run has downloaded and cached the model, the recognition itself needs no connection. The very first conversion does need internet access to fetch the model.