- AI tool overlays translations directly onto images
- Upgrade tackles long-standing gaps in low-resource languages
Qwen, Alibaba’s AI assistant, has rolled out an upgraded image translation feature supporting 119 languages, positioning it among the first tools to offer “image-to-image” translation at near-global linguistic coverage.
The update allows users to upload photos and receive translations directly overlaid onto the original image, preserving layout and formatting.
The company said the feature covers languages used by roughly 98% of the world’s population, including niche tongues such as Faroese, which has fewer than 100,000 speakers.
Handling low-resource languages has long been a bottleneck for AI translation due to limited training data.
Qwen said it addressed recognition and formatting challenges across several such languages, including Georgian and Sardinian, aiming to improve accuracy in both text understanding and visual reconstruction.
Visually aligned
Unlike conventional translation tools that output plain text, the feature maps translated content back onto the original image—allowing menus, prices and annotations to remain visually aligned.
The company has also optimized the tool for 11 travel-related scenarios, including menus, road signs, product packaging and receipts.
The release comes ahead of the peak travel season during the May holiday, as demand grows for real-time translation tools that can handle multilingual, image-based content.
Qwen also launched a user campaign encouraging submissions of foreign-language menus, offering reimbursements for selected entries.
