Image translation
Image translation recognizes text inside webpage images and overlays the translation on the image. It is off by default and can be enabled in settings. Once enabled, hover over an image for about 0.6 seconds to reveal a faint icon in its lower-left corner. The icon becomes clear when you hover over it; click to translate.
The hover entry skips common avatars, icons, logos, emoji, video previews, and small images where it can identify them. Video previews on sites such as X/Twitter can be skipped before playback starts, using player or thumbnail markers. Enlarging a small image may still leave the entry hidden. If an image you want to translate has no icon, use its context menu or open a clear original and try again.
Translate one image
- Make sure image translation is enabled. Click the icon near the image’s lower-left corner, or use the image’s context menu.
- For a new source language, follow the prompt to download its recognition pack and translate.
- Wait for recognition and translation. Choose cancel if you want to stop.
- Switch between original and translated image, or open the text view to copy the full result.
Canceling a translation does not remove downloaded language packs. Later images can reuse them.
If the translation connection is interrupted, it retries once automatically and remains cancelable. If it still fails, choose Retry. After an extension update or reload, refresh the webpage before trying again.
Choose the source language
The recognition pack must match the image’s language. Automatic detection prepares Simplified Chinese, Traditional Chinese, English, and Japanese by default. Other supported languages require selecting the source language and downloading its pack.
The current translation service and target language are used. Hover and context-menu entries can be turned off separately in image settings.
Small or blurry text?
Open a clear original if possible. Vertical text, decorative fonts, complex backgrounds, and tables are harder to recognize. Check names, numbers, and units.
Long translations may appear small inside the picture; use the text view to read them fully. Background repairs may leave marks, and you can always restore the original.
For one small part, try area translation. Unreadable image sources and restricted pages may not work.
What gets sent?
Recognition happens locally. Recognized text goes to the selected translation service; image pixels are not uploaded as part of text translation. Initial language-pack downloads require a network connection. See Data & privacy.