Image to Text (OCR)
Extract text from any image — receipts, IDs, business cards, screenshots, scanned documents. 15 languages. Runs in your browser using the same OCR engine (Tesseract) the cloud services use, so your sensitive documents never travel.
Drop an image to extract text
Receipts, screenshots, business cards, IDs · never uploaded
How OCR works in your browser
ToolChop uses Tesseract.js, the standard JavaScript OCR engine (Apache 2.0 license). On first use it downloads the OCR WebAssembly (~2 MB) and the language pack for your chosen language (~10–15 MB), then caches both in your browser. Every subsequent OCR runs in milliseconds-to-seconds without re-downloading. The image is decoded, passed to the engine, and the extracted text is returned with a confidence score.
Why a local OCR matters
OCR is most often applied to the most sensitive documents people own — receipts (financial), business cards (contact info), IDs (government documents), signed contracts (legal), medical forms (PHI). Uploading these to a cloud OCR service is exactly the kind of data exposure the documents exist to prevent. ToolChop runs the same Tesseract engine locally so the image and the extracted text both stay on your device.
What OCR is good at vs not
- Excellent — printed text (books, articles, signs), screenshots with native text, business cards, receipts with clear printing
- Good — scanned documents (skew-corrected), forms with printed fields
- OK — block-letter handwriting, low-contrast photos of text
- Poor — cursive handwriting, severely skewed documents, stylized fonts, low-light photos
Frequently asked questions
How do I extract text from an image online for free?
Drop your image, pick the language, and click Extract text. ToolChop's OCR engine runs in your browser and returns the recognized text with a confidence score. Copy to clipboard or download as .txt. No account, no upload, no daily limit.
Does ToolChop upload my image?
No. The entire OCR pipeline — image decode, language model, recognition — runs in your browser using WebAssembly. Your image and the extracted text never leave your device. This is the single biggest privacy moat in image processing, because the kinds of images people OCR are almost always sensitive.
Why is the privacy story important for OCR?
OCR is most often applied to receipts (financial data), business cards (personal contact info), driver's licenses and passports (government ID), signed contracts (legal documents), medical forms (PHI), and screenshots with private app content. Uploading these to a third-party OCR site is exactly the kind of data exposure those documents exist to prevent. ToolChop runs OCR locally so the image never travels.
What languages are supported?
15 languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Chinese (Simplified), Chinese (Traditional), Japanese, Korean, Arabic, Hindi, Turkish. Each language requires downloading a one-time language pack (~10–15 MB) which is then cached in your browser.
How accurate is the OCR?
Very accurate for clean, high-contrast text — typically 95%+ for printed documents, screenshots, and business cards. Accuracy drops for handwriting (~50–70%), poor-quality scans, low-light photos, skewed documents, and stylized fonts. The confidence score after recognition tells you how confident the engine is in its result.
Why is the first run slower than later runs?
First run downloads the OCR engine WebAssembly (~2 MB) and the language pack (~10–15 MB). Both are cached in your browser's IndexedDB after the first run, so subsequent images in the same language process in seconds. Switching to a new language re-downloads its pack.
Will OCR work on handwriting?
Tesseract (the engine ToolChop uses) handles printed text very well and handwriting poorly. For neat block-letter handwriting it's ~50% accurate; for cursive it's near-useless. For handwriting specifically, you'd need a different engine — there isn't a great client-side option for that workload yet.
Does it preserve formatting and layout?
Partially. Line breaks and paragraph separations are usually preserved. Multi-column layouts can get scrambled — the engine reads top-to-bottom left-to-right and may interleave columns. Tables are extracted as text but without cell structure. For structured-data extraction (forms, tables), the text output is a starting point, not the final structure.
Can I OCR a PDF?
Not directly here — but you can: first use the PDF to Text tool (extracts existing text from PDFs that have a text layer), or use PDF to JPG to render each page as an image, then drop each image here for OCR. For scanned PDFs without text layers, image-then-OCR is the right path.
How large can the image be?
Tesseract handles multi-megapixel images comfortably. Very large images (above ~20 MP) may slow recognition significantly because the engine scales internally. For best speed, resize images to about 2000–3000 px on the longest edge before OCR — text accuracy doesn't improve beyond that threshold for most fonts.
Why is the confidence number low even though the text looks right?
Confidence is a per-character probability average — short text or unusual characters can pull the average down even when individual words are correct. Conversely, sometimes confidence is high but the text has obvious errors. Always eyeball the extracted text against the source image rather than trusting the score alone.
Why use ToolChop instead of an online OCR service that uploads my image?
Privacy. The kinds of documents people OCR are almost always sensitive — financial, legal, medical, personal ID. Uploading them to a third-party OCR service is the opposite of what those documents exist to protect. ToolChop runs the same Tesseract OCR engine, but locally in your browser. Image and text both stay on your device.