handwriting → editable text

Handwriting OCR

Turn handwritten notes, letters and journal pages into editable digital text. An AI layout model reads cursive and print alike — and the result sits next to your original image so you can proofread and fix it line by line, no retyping.

no photo yetjpg · png · webpstep 1 of 3

up to 20 at once — one file back

no account needed

one MD file, one click away

MDthis is what comes backexample

pipeline

How it works

01Upload the handwritingA photo or scan of notes, a letter, a journal page or a whiteboard.
02AI recognitionA vision-language model reads the writing — far beyond classic print-only OCR.
03Proofread & copyFlip to the original image to verify, then copy or export the text.

why this tool

Handwriting OCR, done right

Reads cursive and printPowered by a vision-language OCR model, not a print-era engine.

Proofread viewToggle between the recognized text and your original image to catch misreads.

More than copy-pasteDownload as .docx or .md, send to Google Docs, or paste into Notion.

Batch up to 20 pagesA whole notebook section in one pass, merged into a single document.

why handwriting is different

Why print OCR fails on handwriting, and what is used instead

Classic OCR was built for printed type: it expects every "a" to look like every other "a", sitting on a straight baseline with clean gaps between letters. Handwriting breaks all three. Letters join, slant and change shape from word to word, lines drift up the page, and a word is often recognisable only from the words around it.

So this tool does not use a print engine. It uses a vision-language model — one that looks at the whole page and reads it the way a person would, using the shape of the line and the sense of the sentence together. That is what lets it read joined-up writing that a character-by-character engine turns into noise.

Print and cursive: Block capitals, neat print and most joined-up handwriting are read. Very fast scrawl and heavily stylised lettering degrade, as they would for a person who does not know the writer.
Page structure: Headings, paragraphs and lists on the page come back as Markdown structure, so a set of notes keeps its shape instead of collapsing into one paragraph.
Crossings-out and margins: Struck-through words and notes squeezed into a margin are the hardest part of any page; expect to touch them up in the proofread view.
Transcription, not rewriting: The aim is what is written — abbreviations and all. No second pass summarises, tidies or rephrases the text.

proofreading

Checking the result against your page

No handwriting recognition is right every time, and the expensive mistake is the one you do not notice. The result is built for checking: the recognised text and your original photo sit in the same panel, one tab apart, so you can read a passage, flip to the photo, and flip back.

Original tab: Your photo, full width in its own scrolling frame — large enough to read the writing against the text line by line.
Preview and source: The rendered text, and the raw Markdown behind it if you want to fix something before copying.
Where to look first: Names, numbers and words you would not expect in the sentence. A model that reads by context gets ordinary words right and is most tempted on the unusual ones.
Then send it on: Copy the text, download .docx or .md, open it in Google Docs, or paste it into Notion.

before you upload

Photographing handwriting so every word survives

Light from the side, no shadow: Your phone's shadow across the page is the most common reason a line goes missing. Light from a window to one side, not from above you.
Pencil needs more light: Pencil is low contrast to begin with. Brighter, even light helps far more than a higher-resolution photo.
One page per image: A two-page spread is read as one page. Photograph each page on its own and drop them together.
Fill the frame: Crop out the desk and the notebook cover — the more of the image the writing occupies, the more detail every word gets.
Your image: It is deleted within 30 minutes of the recognition and the job record within two hours. Nothing is kept to train a model.

faq

Frequently asked questions

Can it read cursive?

Yes — the AI model handles print and most joined-up handwriting. Very messy scrawl or heavily stylized writing will degrade; the built-in original-image view makes those misreads quick to spot and fix.

How accurate is it with messy handwriting?

Accuracy depends on legibility and photo quality. Neat handwriting on a clean, well-lit page is usually near-perfect; for harder pages, proofread against the original view — you edit the text, not retype the page.

What can I do with the recognized text?

Copy it, download it as Markdown or .docx, export to Google Docs, or paste into Notion. Multiple pages uploaded together merge into one document.

How many pages can I convert without signing in?

A few a day. Handwriting goes to the larger vision model rather than a print engine, so each page costs more to read, and pages without an account are capped lower than plain text extraction. Signing in is free during the beta and raises the limit — and unlocks dropping a whole notebook section at once.

Will it fix my spelling?

No — nothing is run over the text afterwards to correct or rephrase it. The aim is a transcription of what is on the page, because one that silently rewrites you is one you can no longer check line by line. Proofread unusual words against the original.

Does it work on a whiteboard photo?

Yes. Shoot the board straight on, close enough that it fills the frame, and away from the reflection of the ceiling lights — glare is what loses words on a whiteboard, not the handwriting.

Can I convert a whole notebook?

Photograph one page per image and drop up to 20 at once on a free account. They merge into one document, with a section per page named after the photo, in the order you dropped them.

next door