Image OCR — Extract Text
Extract text from a photo, cancel recognition, correct the result and download TXT locally.
Recognition runs in your browser; your image and extracted text are not uploaded. OCR software and language models download from public CDNs, which receive normal connection details. Review the result before relying on it.
Recognize printed text in a local PNG, JPEG or WebP, select one of twelve languages, edit the result and copy or download TXT.
How to use it
- 1Choose an image within the 10 MiB and 12-million-pixel limits.
- 2Select its text language and choose Recognize text.
- 3Review and correct the recognized text.
- 4Copy it or download the edited text as TXT.
Example
- Input
- Clean PNG reading HELLO TOOLVORE; English model
- Output
- HELLO TOOLVORE
Verified fixture, not an accuracy guarantee for other images.
What happens to your data
The image and recognized text stay on your device. OCR downloads Tesseract engine and language assets from jsDelivr when needed; offline availability is not guaranteed.
Last updated October 2026
Extract editable text from a local screenshot, scan or photograph of printed words. Open a PNG, JPEG or WebP image, select the text language and choose Recognize text. Review the result in the editable box, correct any mistakes, then copy it or download a plain-text file.
OCR is useful when words are visible but cannot be selected: an instruction in a screenshot, a printed note or a scanned page. It estimates characters from pixels, so a plausible-looking result can still contain a wrong number, missing accent or reordered line. Keep the original image beside the text while proofreading it.
How it works
The tool checks file signatures and header dimensions before asking the browser to decode an image, then checks its decoded dimensions again. Inputs are limited to 10 MiB, 12 million pixels and 8,192 pixels per side. SVG, GIF, HEIC and PDF are not accepted. Animated PNG or WebP is treated as a browser-decoded still frame. The browser-oriented image is drawn onto a white background for recognition, so transparency is flattened.
Recognition starts only when you choose Recognize text. Tesseract.js 7 runs in browser workers; your selected image and recognized text are not sent to an OCR server. The engine and selected language model are downloaded from jsDelivr. These are network requests for program and model files, not uploads of your image. Download sizes vary by model and browser capabilities. Cached model data may make later runs faster, but this page does not promise that OCR will work offline.
One recognition job runs at a time. Cancel, replacing the image, changing the language or leaving the page stops its worker. A job also stops after two minutes; a smaller image or faster connection may help. Changing image or language clears the previous text so that it cannot be mistaken for the current image's result. Choose Recognize text again after a language change.
The language menu offers twelve individual models, including English, Hindi, Arabic, Simplified Chinese, Japanese and Korean. Selecting a model is not automatic language detection, and mixed-language pages may need separate passes. Printed, upright text with clear contrast is the best starting point. This is not a dependable handwriting recognizer, a table-to-spreadsheet converter or an exact document-layout reconstruction tool. TXT downloads contain the current edited text in UTF-8, without fonts, images or page positioning.
Worked example: recognize a clean printed line
- Input
- A 1,000 × 240 PNG with black text HELLO TOOLVORE on white; English
- Result
- HELLO TOOLVORE
Verified on this clean test image. Other fonts, photos and layouts can produce different accuracy.
Common use cases
- Copying an instruction from a screenshot
- Recovering printed words from a local scan
- Creating an editable first draft from a photographed note
- Saving corrected recognized text as a small UTF-8 TXT file
Frequently asked questions
How do I turn a picture into text?＋
Choose a supported image, select the language printed in it and choose Recognize text. Wait for the editable result, proofread it, then copy it or download TXT. Selecting an image alone does not start model downloads.
Is my image uploaded for OCR?＋
No. Recognition runs on your device. The browser does contact jsDelivr for engine files and language data when needed; the selected image and recognized text are not included in those asset requests.
Why does the first recognition take longer?＋
The engine and selected language model may need downloading and initialization. Their sizes vary. Model data may be cached in browser storage. The progress line reports the current stage, and Cancel stops the current job.
Can it read handwriting or preserve a table?＋
Neither is reliable here. The output is plain text, so table structure, columns and handwriting can be misread or reordered. Use a clear upright crop of printed text, and check every important value against the source.
What happens if I change the language?＋
Any active recognition stops and old text is cleared. The image stays selected. Choose Recognize text to run the newly selected model; the tool does not silently rerun or combine languages.
Why did OCR return no text?＋
The image may have no readable words, the wrong model may be selected, or the words may be too small, blurry or low contrast. Try a clearer crop with some margin around the text. An empty result is different from an engine download error.
Related tools
Extract Text from PDF
Pull the selectable text out of a PDF to paste elsewhere. It reads the text layer, so scans give nothing.
Text to PDF
Create A4 or Letter PDFs with checked font coverage, wrapped text, titles and local TXT import.
Crop Image
Trim a photo to the part you want, freehand or locked to a ratio like 1:1 or 16:9.
Image Compressor
Compress JPG, PNG and WebP images to a smaller file size — or to an exact target size like 100 KB.