Handwriting to Text Converter
Transcribe handwritten notes and forms into digital text.
Best for digitizing handwritten notes, filled forms, and letters you need to search or edit.
▸▾English handwriting model, ~262MB on first use
Uses Florence-2 (full-page) then TrOCR (line-by-line) for cursive and mixed handwriting. English only — other languages use a print-oriented fallback. Downloads ~262MB once (shared with the Image Description tool, cached after that). On iPhone Safari the model is skipped to avoid a tab crash and the fallback engine runs instead. For English output, "Fix common OCR errors" applies a dictionary pass to catch classic mistakes like rn→m or O→0 — only low-confidence words are touched, shown in the review panel so you can revert before downloading.
Drop files or a folder here
or click to browse · paste from clipboard
Accepts .JPG, .JPEG, .PNG, .WEBP, .HEIC, .HEIF · Up to 1,000 files
How it works
Drop your files
Drag and drop, click to browse, or paste from clipboard. Up to 1,000 files at once.
Choose settings
Adjust quality, format, and other options to match your needs.
Click Convert
Everything runs in your browser via WebAssembly. Handwriting to Text Converter happens locally — no server involved.
Download
Download files individually or grab all at once as a ZIP.
Frequently asked questions
No. The handwriting model runs entirely in your browser. Your files never leave your device.
The first conversion downloads a ~262MB handwriting model (Florence-2 + TrOCR), then caches it in your browser. Later conversions skip the download. The model is shared with the Image Description tool, so it may already be cached.
Yes, for English. The model is trained on handwritten input, including joined strokes. Messy or doctor-style scrawl will still produce errors — treat the output as a first draft.
Handwriting is the hardest input for OCR. Neat, upright print on a white background can reach 90–95% accuracy. Casual cursive, mixed styles, or anything on a coloured or patterned background will be lower. The review panel underlines low-confidence words in amber so you can focus corrections quickly.
Yes. Printed form labels extract cleanly. Handwritten answers in the blanks extract with variable accuracy — review those fields before downloading.
Yes. After conversion, a review panel shows words flagged with low OCR confidence underlined in amber. Auto-corrected words have a blue dotted underline — click to see the original and revert. Click "Apply changes" to lock edits before downloading.