Choose an image, then click Extract text; the recognized text appears below once the scan finishes.
Image to Text (OCR)
Pull the text out of a photo, screenshot, or scanned page without installing anything or sending the image anywhere. The image opens in this browser tab, the recognition engine runs locally, and the recognized text lands in a box you can copy or download as a .txt file - the file itself never leaves your device.
Works best on clear, printed text with good contrast - accuracy drops on handwriting, low-resolution photos, or heavily skewed or rotated text. Recognition covers one language at a time (English by default) and returns plain text only - it does not preserve the original layout, columns, or fonts, and does not return bounding-box coordinates for the recognized words.
The first run on a device downloads the recognition engine and language data (a few megabytes) before it can scan anything, even though the image itself is never uploaded; the browser caches that download so later runs on the same device start faster.
Frequently Asked Questions
Does the image get uploaded anywhere?
No. The image opens in this browser tab and the recognition runs locally - it is never sent to a server. The only network activity is a one-time download of the recognition engine and language data on the first run on a device (a few megabytes); later runs on that device are faster because the browser caches it.
Does it work on handwriting?
Not reliably. The recognition engine is trained on printed text, so results are best on clear, printed text with good contrast. Accuracy drops on handwriting, low-resolution photos, and heavily skewed or rotated text.
Does it keep the original layout, like columns or tables?
No. The output is plain recognized text only - it does not preserve the original layout, columns, tables, or fonts, and does not return bounding-box coordinates for the recognized words.
Can it recognize more than one language at once?
It recognizes one language at a time (English by default) and does not auto-detect the language in the image, so mixed-language images will misrecognize the portions in the non-selected language.