Image OCR turns printed text in screenshots, photos and scans into text you can edit and copy. The tesseract.js engine runs straight in the browser: drop in several images at once to build a queue, or press Ctrl+V to paste a screenshot you just took. Every finished image reports a confidence percentage, and the result box lets you fix a typo before you copy or download.

It works on images only. For a scanned PDF use “PDF OCR”, which renders page by page and can output a searchable PDF; to compare two engines, “Dual-engine OCR” goes through an online endpoint and has different privacy properties.

1. 01Pick the recognition languagethe default is “Simplified Chinese + English (recommended)”, with “English only”, “Traditional Chinese” and “Japanese” alongside it. The Chinese and English language data ships with the site; Traditional Chinese and Japanese are downloaded on first use (about 2–20 MB) and then cached.
2. 02Add imagesdrag and drop, pick from the file dialog or press Ctrl+V to paste a screenshot; JPG, PNG, WebP, BMP, GIF and TIFF are accepted, several at a time, and they form the “image queue”.
3. 03Start recognitionclick “Start recognition” and the progress bar walks through “image x of n · loading recognition engine / downloading language data / recognising text”. The first image has to load the engine, and the ones after it reuse the same process and go much faster. “Cancel” is available at any point.
4. 04Review and exportclick a finished image in the queue and “Recognition result” shows the text with “confidence xx%”; edit it, then “Copy” or “Download TXT”. With several images, “Download all results as a ZIP (n TXT files)” produces OCR-results.zip, adding (1), (2) and so on to duplicate names.

Input formats

JPG, PNG, WebP, BMP, GIF and TIFF; anything that is not an image is ignored.

Languages

only Simplified Chinese + English, English only, Traditional Chinese and Japanese; when the language does not match the text in the image the output is garbage or blank, and switching and re-running fixes it.

First run

the engine and the language data have to be loaded the first time (the engine and the Chinese/English data come from this site, Traditional Chinese and Japanese from a public CDN), which can take a dozen seconds; changing language rebuilds the engine.

Accuracy

it depends on sharpness, contrast and typeface, and clean printed text is best; handwriting, decorative lettering, vertical text, busy backgrounds and tilted photos drop off sharply, so always proofread.

What it does not do

it does not recognise PDFs (use “PDF OCR”), does not keep the layout or table structure (plain text only) and does not translate; straighten a crooked photo with “Scan Enhance” before recognising it.

Images are recognised in the browser and are never uploaded to any server. The engine and the Simplified Chinese/English language data are self-hosted by this site and downloaded on first use; choosing Traditional Chinese or Japanese downloads language data from the tesseract.js default public CDN. Those requests fetch engine and language files only, never the content of your images.