Image to Text Converter — Extract Text from Images & PDFs
Turn photos, scans, screenshots and PDFs into editable text with real OCR that runs entirely in your browser — nothing you add is ever uploaded. Read printed text in 102 languages, including Bengali and mixed বাংলা + English documents in a single pass, in batches of up to 20 files, with per-word confidence review and TXT, DOCX and ZIP exports. No signup, no page limits, no watermarks.
Drop images or PDFs here, or click to choose files
JPG, PNG, WebP, GIF, BMP, AVIF or PDF · up to 20 files · 25 MB each · or paste a screenshot with Ctrl+V / ⌘+V
Real OCR, on your device — nothing is uploaded
Recognition runs the open-source Tesseract OCR engine (WebAssembly) inside your browser. The engine and the language pack you pick download once from public CDNs (~2–4 MB per language) and are cached by your browser; your images and PDFs themselves never leave this device — there is no upload, no account and no usage limit. Printed text reads best; handwriting is honestly hit-and-miss.
How to Use
- 1Add up to 20 images or PDFs — drag and drop them, browse, paste a screenshot with Ctrl+V / ⌘+V, or try the built-in sample document.
- 2Pick the language of the text (Bengali and English are pinned at the top). For mixed documents, add a second language or use the বাংলা + English quick-set.
- 3If the photo is faded, small or sideways, turn on Enhance contrast or Black & white, rotate it, and choose a layout mode — then click Extract.
- 4Check the confidence score, use Review to see the exact words the engine was unsure about, edit anything, then copy or download TXT, DOCX or ZIP.
Key Features
- ✓
Real OCR that never uploads your files
Most image-to-text sites make you upload your document to a stranger's server and wait in a queue. This one runs the open-source Tesseract OCR engine — the same engine Google maintains — as WebAssembly inside your browser. Your receipts, certificates and contracts are read on your own device and never transmitted anywhere. There is no account, no hourly cap and no page counter, because there is no server doing the work.
- ✓
Bengali first-class — and mixed বাংলা + English in one pass
102 recognition languages, with Bengali pinned to the top of the list instead of buried in it — and a one-tap বাংলা + English preset for the documents Bangladesh actually produces: certificates, office memos and forms that mix both scripts on the same page. Add any second language to any first, and mixed pages are read in a single pass instead of two error-prone ones.
- ✓
A batch workspace, not a one-file form
Drop up to 20 images or PDFs at once — or paste screenshots straight from the clipboard — and watch each file move through the queue with its own progress, word count and confidence score. PDFs are rendered and read page by page in your browser (up to 20 pages). When the batch is done, take everything as one combined file, a Word document, or a ZIP of individual texts.
- ✓
Confidence you can act on
Every result carries the engine's real confidence score — and a Review view that highlights the exact words the engine was unsure about, with their scores listed. Instead of proofreading every line, you check the dozen words that are actually likely to be wrong. It is the difference between a number that looks reassuring and information that saves you time.
- ✓
Preprocessing that rescues bad photos
Phone photos of documents fail OCR for predictable reasons: low contrast, small text, sideways orientation. This tool fights all three — automatic upscaling for small screenshots, a contrast-stretch Enhance mode, an Otsu black & white mode for faded print, one-click 90° rotation for sideways shots, and plain-English layout modes (full page, single block, single line, sparse text) for receipts, signs and forms.
- ✓
Exports that fit real work
Copy with one click, or download .txt for notes, an editable .docx for Word and Google Docs, a combined file for a whole batch, or a ZIP when each source needs its own text file. Results land in an editable box first, so a stray character can be fixed before the text goes anywhere.
Frequently Asked Questions
Is this image to text converter really free?
Are my images or PDFs uploaded anywhere?
Which languages can it read — and what about Bengali?
Can it extract text from a PDF?
How accurate is it, and how do I improve a bad result?
Can it read handwriting?
Why is the first use slower than later ones?
What formats and sizes are supported?
What "image to text" actually involves
An image of text is just coloured pixels — a photograph of a page does not contain letters a computer can search, copy or edit. Optical character recognition (OCR) is the process that bridges the gap: the software finds the shapes that look like characters, decides which letters they are, and rebuilds them as real text. This page does that with Tesseract, the open-source OCR engine originally built at HP, maintained by Google for two decades, and used everywhere from library digitisation projects to the Internet Archive. Tesseract's recognition core is an LSTM neural network — that is the honest extent of the "AI" in tools like this one, and it is why this page describes itself as OCR rather than dressing the same engine up as something magical. The engine has been compiled to WebAssembly, so the exact software that powers desktop OCR runs inside your browser tab at close to native speed.
Why running it in your browser matters
The classic online OCR flow is: upload your file, wait for a server, download the result. It works, but it has three costs people rarely count. Privacy: the documents people most need converted — invoices, medical reports, legal papers, ID-adjacent forms — are exactly the ones that should not be emailed to an unknown server, however sincere its deletion policy. Limits: server time costs money, so upload tools ration it — five files an hour for guests is a typical cap, with sign-up walls behind it. Speed: a big scan spends longer travelling than being read. Running the engine locally removes all three: nothing is uploaded, so there is nothing to ration and nothing to leak. The only downloads are the engine itself (a few megabytes, once) and the language pack for the language you pick — both cached by your browser, which is also why the tool keeps working offline once installed as an app. If your next step after extracting is shrinking the source photos for sharing, the image compressor keeps the same nothing-uploaded rule.
Bengali, and the mixed-document problem
OCR language models are script-specific: an English model shown a Bengali page does not produce Bengali with errors, it produces confident nonsense, because it maps every glyph to the nearest Latin shape. That is why the language picker matters more than any other control on this page — and why Bengali (বাংলা) sits at the top of it rather than at position forty in an alphabetical list. The subtler problem is the mixed page. A Bangladeshi university certificate, a trade licence, an office memo: Bengali prose with English names, dates and registration numbers woven through it. Run a single-language model and one of the two scripts is always wrong. The বাংলা + English quick-set loads both models together so the whole page is read in one pass — the same trick works for any pair, from Hindi + English to Arabic + French. In total, 102 languages are one click away, each downloaded once (roughly 2–4 MB) and cached for every future visit.
Getting a good read out of a bad photo
When OCR disappoints, the cause is almost never the engine — it is the photograph. Four fixes cover most failures, and all four are built in. Resolution: text rendered too small starves the recognizer, so small screenshots are automatically upscaled into the size band the model was trained on, while enormous camera frames are stepped down to what a browser can hold. Contrast: Enhance mode stretches the photo's tonal range (faded ink on grey paper is the classic failure), and Black & white mode applies an Otsu threshold that separates ink from paper decisively — often the single biggest win on receipts and old scans. Orientation: the rotate button turns a sideways phone photo upright in 90° steps; OCR has no tolerance for rotated text. Layout: the layout mode tells the engine what shape of text to expect — a full page, one block (receipts, screenshots), a single line (labels, signs), or sparse text scattered anywhere (menus, whiteboards, forms). Add straight, even lighting at capture time and printed text in any of the 102 languages reads at the accuracy the confidence score then reports honestly.
Confidence scores, and checking only what needs checking
Tesseract scores every word it reads. Most tools either hide that or average it into a single feel-good percentage. Here the average is shown per file — and, more usefully, the Review unsure words view highlights the individual words that fell below 60% confidence directly in the text, with the weakest listed underneath. A page that scores 97% might still misread one account number; the review view points at exactly the suspects so a thirty-second check replaces a five-minute proofread. Treat any low-confidence number, date or name as unverified until your eyes have seen it — that habit matters more than the engine choice. If the text you extract is destined for a document or a web page, the Markdown to HTML converter and the word counter pick up where this tool leaves off.
The honest limits
Three, stated plainly. Handwriting: the models are trained on printed and typed text; neat block capitals sometimes come through, ordinary handwriting mostly does not — any tool that promises otherwise for free is overselling. Layout: you get faithful plain text in reading order, not a pixel-perfect recreation of the page — complex tables and multi-column designs may need tidying, which is why results arrive in an editable box and export to Word. First-run downloads: the first time you use a language, its pack downloads before recognition starts; the progress bar names that stage honestly instead of pretending the engine is "thinking". After that one download, the same language starts instantly — and because everything after the download happens on your device, a confidential document is exactly as private as the laptop or phone you are holding.
Related Tools You Might Like
Handy utilities in Image Tools
Word Counter
Count words, characters, sentences and reading time live as you type — with goals, platform limits and keyword density. 100% private.
Image Compressor
Compress JPG, PNG and WebP images online for free — by quality slider or to an exact size like 100 KB. Batch ZIP download. No uploads, no sign-up.
Markdown to HTML Converter
Convert Markdown to clean, sanitised HTML with a live preview, GFM tables, footnotes, alerts and export themes. Free, private, in your browser.