01 / Read
The text is found
Pick the language the text is written in. A text reader in your browser finds each line and groups lines into blocks, such as one sign or one menu item.
Free tool
Drop a photo of a sign, a menu, a label or a screenshot. The page reads the words, translates them, covers the old ones and draws the new ones where they were. You get the translated picture to save, a list of the text to copy, and a box next to every line to fix a wrong reading or a wrong translation. The picture is never uploaded.
Text list Fix the reading or the translation and the picture updates. Untick a block to leave it as it was.
Look of the picture
Some blocks were added or switched on since the AI ran. Tick the box off and on to include them. Until then they use the colour fill.
Your picture is read and translated in your browser. It is never uploaded.
How it works
01 / Read
Pick the language the text is written in. A text reader in your browser finds each line and groups lines into blocks, such as one sign or one menu item.
02 / Translate
Chrome on a computer uses its built-in translator. Other browsers download an open translation model once, for a shorter list of languages.
03 / Draw
The old words are covered by the colours around them, or rebuilt with an AI model. The translation is sized to fit and drawn in a readable colour.
Guide
An image translator finds the text inside a picture, translates it, and puts the translation back. This one runs in your browser, so a photo of a letter, a menu or a screenshot of a chat never goes to a server. For an image translator to English, choose the language of the text, leave Translate to on English, and drop the picture. You get the translated picture, a list of the text, and a way to fix mistakes before you save.
The translation is done on your computer, by one of two engines, picked for you.
| Your browser | Engine | Pairs | First use |
|---|---|---|---|
| Chrome on a computer | Chrome's built-in translator | Many pairs among the languages in the menu | Chrome may fetch a small language pack once |
| Other browsers | Open translation models | Into English from Spanish, French, German, Italian, Dutch, Portuguese, Russian, Arabic, Hindi, Indonesian, Japanese, Korean, Chinese, Turkish and Bengali; from English into Spanish, French, German, Italian, Dutch, Russian, Ukrainian, Arabic, Hindi, Indonesian, Chinese, Vietnamese and Portuguese | A model of about 115 MB downloads once for each pair |
With the downloaded models, a pair without English in it, such as Spanish to German, goes through English in two steps. Each step can shift the meaning, so read those results with extra care. When your browser cannot do a pair, the Translate to menu only lists the pairs it can.
Two ways, and you can switch between them with one tick box.
Whichever you pick, the text list and the plain text download do not depend on it. If the AI model is slow or cannot start on your computer, press Stop the AI and keep the colour fill.
It reads horizontal printed text. It does not read upright (vertical) writing, handwriting, text on a curve, or very small, blurred or heavily decorated lettering. Letters on a busy background are read less well than letters on a plain one. The new text uses your computer's fonts, so it will not match the original lettering, and logos are not translated. A machine translation helps you understand a picture. For legal, medical or safety information, ask a person who knows the language.
To get the words only, see Image to text. For a document, use PDF translator.
Questions
No. The picture is read, translated and redrawn in this tab. The reader and its language data are served from this site. The translation model, when your browser needs one, and the optional AI background model are downloaded to your browser from Hugging Face, the open model library, and they carry only the model, never your picture.
Drop the picture, set Text in the picture to the language of the words, and keep Translate to on English. The translated picture appears next to the original.
Arabic, Bengali, Chinese (Simplified), Dutch, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Russian, Spanish and Turkish. You can translate them into the languages in the second menu.
Check Text in the picture. The reader needs the language of the words, since it cannot guess it. If it is wrong, choose the right one and press Read the picture again.
Yes. Every block has two boxes in the Text list, one for what was read and one for the translation. Type in either and the picture redraws.
No. It reads horizontal lines only. For upright text, type the words into the Text list by hand, or drag a box and fix the reading.
No. The translation is drawn in a font from your computer, sized to the space, in a colour taken from the original or black or white, whichever is easier to read.
The reader's language data for your language, about 1 to 3 MB, from this site. In Chrome, its own language pack may download once. In other browsers, a translation model of about 115 MB per pair. The AI background option adds 208 MB, once. Your browser keeps them for next time.
Yes. Paste it with Cmd+V or Ctrl+V. Screenshots have flat backgrounds, so the colour fill works well. Switch Line breaks to Keep every line if each message should be translated alone.
It is free, with no account and no daily limit set by this page. The pictures are scaled down to 4000 pixels on the longest side for speed and memory.