Image Translator

No. The OCR engine, its WebAssembly core and the English language pack are served from this site and run inside your browser, so the image is read on your own machine. The text that comes out of it is then sent to a translation model, because translating needs one. The picture itself never leaves your computer.

English, and other things written in the Latin alphabet. That is the only language pack this page ships — it is about 4 MB, and shipping one per script would be tens of megabytes for every visitor. Text in another script comes back as nonsense, which you will see in the list under the result before you trust anything.

Nineteen, including Simplified and Traditional Chinese, Japanese, Korean, Spanish, Portuguese, French, German, Italian, Russian, Arabic, Hindi, Indonesian, Thai, Vietnamese, Turkish, Dutch and Polish. Pick from the list on the tool.

Because that is what it is. The original text is covered with the average colour found around it, and the translation is drawn on top in a plain system font. On a white document or a flat screenshot the patch is invisible. On a photo, a gradient or a busy background you will see it. Fonts, colours, bold, italics and any effect on the original text are not reproduced.

OCR wants flat, sharp, reasonably large text. Handwriting, text at an angle, curved text on a bottle, low resolution, heavy shadow or a photo taken from the side all produce wrong words or nothing. When nothing readable is found the tool says so and stops — nothing is sent and none of your allowance is used.

Usually. A translation is often longer than its source, so the type is shrunk to fit the box it belongs to. When even the smallest size does not fit, the text runs past the edge of its box rather than being cut off, because a truncated translation is worse than an untidy one.

A PNG at the same pixel size as the picture you put in. No watermark, no sign-up, no waiting queue.