Free OCR Tools — Extract Text from Images & PDFs
OCR (optical character recognition) turns a picture of text into text you can actually select, copy and edit. These tools do it without uploading anything: the recognition engine runs inside your browser, so a payslip, an ID scan or a contract never leaves your device. Thirteen languages are supported, 8 of them Indian scripts.
Which OCR tool should I use?
Both use the same recognition engine. The difference is what you feed them.
| If you have… | Use | Why |
|---|---|---|
| A screenshot or photo | Image to Text | Takes JPG, PNG, WebP and HEIC directly — no conversion step. |
| A scanned PDF | PDF OCR | Renders every page and runs recognition across the whole document in one pass. |
| A PDF whose text is already selectable | PDF to Word | You do not need OCR at all — the text is already there, and extracting it directly is both faster and perfectly accurate. |
| A photo that is hard to read | Image Filters first | Raising contrast before OCR usually beats any post-processing. |
What this does that most free OCR sites don't
It rebuilds layout, not just text
Most tools return one undifferentiated block of text. This engine uses each word's position on the page to reconstruct headings, bullet lists and tables — so a scanned invoice comes back as a table you can paste into a spreadsheet.
It shows you what it wasn't sure about
Every recognised word carries a confidence score. Words the engine was unsure of are highlighted, so you can proofread the three words that need it instead of re-reading the whole page.
It exports somewhere useful
Results download as Word, HTML, Markdown or plain text — and the PDF tool can rebuild a searchable PDF with an invisible text layer over the original scan.
Nothing is uploaded
This matters most for exactly the documents people OCR: ID cards, bank statements, medical letters. Open your Network tab and watch — the engine downloads, your file does not upload.
Supported languages
Select the document's language before running. This matters more than people expect: recognition is script-specific, and running a Devanagari page as English produces near-total garbage rather than slightly worse output.
| Language | Script | Region |
|---|---|---|
| English | Latin | International |
| Hindi | Devanagari | India |
| Bengali | Bengali | India |
| Odia | Odia | India |
| Tamil | Tamil | India |
| Telugu | Telugu | India |
| Marathi | Devanagari | India |
| Gujarati | Gujarati | India |
| Punjabi | Gurmukhi | India |
| Arabic | Arabic | International |
| French | Latin | International |
| Spanish | Latin | International |
| German | Latin | International |
How to get accurate OCR results
OCR accuracy is decided almost entirely by the input. In roughly descending order of impact:
Scan flat, don't photograph at an angle
Perspective distortion is the single biggest killer of accuracy. A page photographed from an angle has letters that skew progressively across the line. If you must use a phone camera, hold it parallel to the page and fill the frame.
Aim for 300 DPI or a wide screenshot
Below roughly 200 DPI, letterforms start to merge and the engine guesses. If you are scanning, 300 DPI is the sweet spot. Going far above 600 DPI mostly costs memory without improving results.
Get contrast up and shadows out
Black text on white is ideal. Grey text on grey, a shadow falling across half the page, or a photo taken in warm indoor light all reduce accuracy. Raising contrast before OCR often fixes a bad result outright.
Straighten the page
Even a few degrees of rotation hurts, because line detection assumes roughly horizontal text. Crop and rotate so the lines run flat before running recognition.
Match the language to the script
Set the language selector to the actual language on the page. Mixed-script documents will always have one script recognised worse than the other — run them twice if accuracy matters.
Then proofread the highlighted words
Words the engine flagged as low-confidence are marked in the output. Those are where the errors are concentrated — checking them is far more efficient than re-reading everything.
Extracted text you need to tidy up? The text tools can strip stray line breaks and collapse extra spaces — both very common after OCR. If the scan also carries metadata you would rather not share, remove its EXIF data first.
Frequently asked questions
Yes. There is no account, no daily page limit and no watermark. The recognition engine downloads into your browser the first time you use it and then runs on your own device.
No. Both OCR tools run entirely in your browser using WebAssembly. Your image or PDF is read from disk into memory on your device and never transmitted. You can confirm this by opening your browser's Network tab while the tool runs — you will see the engine download, but no upload of your file.
Thirteen: English, Hindi, Bengali, Odia, Tamil, Telugu, Marathi, Gujarati, Punjabi, Arabic, French, Spanish and German. Pick the document's language before running — accuracy drops sharply if the selected language does not match the script on the page.
It reconstructs structure. The engine reads each word's position on the page and rebuilds headings, bullet lists and tables rather than returning one flat block of text. You can export the result as Word, HTML, Markdown or plain text.
Not reliably. The engine is trained on printed type. Neat block capitals sometimes work, but cursive and casual handwriting will produce poor results. This is a genuine limitation, not a setting you can change.
Almost always resolution or contrast. A photo of a page taken at an angle in dim light is the worst case; a flat, well-lit scan at 300 DPI is the best. See the accuracy checklist above — following it typically matters more than any option in the tool.
There is no artificial cap. The practical limit is your device's memory, since the whole job runs locally. Very long scanned PDFs on an older phone may run out of memory — split them first if that happens.
Explore more free tools
Every category runs free in your browser — nothing is uploaded.