Private. In-browser. No upload.
Pull text out of any image — without uploading it.
Extract typed text from screenshots, photos of printed pages, scanned receipts, slides, and signs. Drop the image, click extract, and you get clean editable text to copy or download. Everything runs on your device — the image never goes to a server. Especially useful when the source matters: tax docs, password resets, prescriptions, anything you'd rather not hand to a stranger.
First-time use downloads the OCR engine (~5MB) and English language data (~5MB). Cached after that — re-runs are instant.
Text extracted
0 words 0 chars 0% confidence
- No upload, ever. The image stays in your browser. Tesseract.js runs as WebAssembly on your CPU.
- Editable result. The recognized text is in a regular textarea — clean it up, copy what you want, or download the whole thing as .txt.
- Works offline after first load. The OCR engine and language model cache locally. Once loaded, you can disconnect from the internet and keep using it.
How it works
- You drop an image. JPG, PNG, WebP, BMP, GIF, or TIFF — up to 20MB. A preview appears immediately.
- Click "Extract text". Your browser loads the Tesseract OCR engine (~5MB) and English model (~5MB) on first use only; cached thereafter.
- The text appears in an editable box. Clean up any minor recognition errors, then copy to clipboard or download as a .txt file.
Questions
- Does my image ever leave my device?
- No. The image stays in your browser the entire time. Tesseract.js — the OCR library — runs as WebAssembly on your device's CPU. We never see the picture. Other online OCR tools upload your image to their servers; this one doesn't. The only network traffic during use is the one-time download of the OCR engine and English language model from a public CDN (jsdelivr), which is just the recognition code itself, not your image.
- What kinds of images work well?
- Screenshots of webpages, app interfaces, error messages, code snippets, terminal output — these work nearly perfectly because the text is sharp and the background is clean. Photos of printed pages, business cards, signs, and receipts work well if the photo is in focus and reasonably well-lit. The text should be upright (not rotated 90°) and at least roughly horizontal. Skip blurry photos, heavily skewed shots, and handwriting — Tesseract was built for printed/typed text and is poor at handwriting.
- Why is the first run slow?
- On first use, your browser downloads the Tesseract OCR engine (~5MB of WebAssembly) and the English language model (~5MB). After that, both are cached by your browser indefinitely, and subsequent runs are nearly instant — recognition itself takes 1–5 seconds for most images. So the first image you process takes maybe 20 seconds total; every image after that is much faster.
- How accurate is the recognition?
- Very good for clean printed text — typically 95%+ accurate on screenshots. Photos are more variable: a well-lit photo of a printed page gets 85–95%, a phone snapshot of a receipt under fluorescent light might get 70–85%. The confidence score shown next to the result is a rough indicator. If the text comes out garbled, try a higher-resolution photo, better lighting, or a less skewed angle. Tesseract is open-source and doesn't beat the very best commercial OCR engines (Google Cloud Vision, AWS Textract) — but those require uploading your image. This is the best in-browser option that exists.
- What languages are supported?
- Currently English only in this tool. Tesseract itself supports 100+ languages — we'll add a language picker in a future update. If you have a specific language need, mention it via the thumbs-down feedback button.
- Can I OCR a PDF here?
- Not directly in this tool. For now, take a screenshot of the PDF page (or use our PDF to Images tool to extract pages as JPGs), then run those images through here. A direct PDF OCR tool is on the roadmap.
- What about handwriting?
- Tesseract handles printed/typed text — not handwriting. Handwriting recognition is a fundamentally different problem (requires neural network models trained specifically for cursive/print variation) and isn't supported here.
- Is the recognition output editable?
- Yes. The result text area is fully editable — clean up any recognition errors before you copy or download. OCR is rarely 100% perfect; we expect you to fix the small mistakes manually. The confidence score gives you a hint about which results need more cleanup.