Building a multimodal image-translation training dataset with a machine translation company

Twigfarm, the company behind LETR AI (formerly LETR WORKS), is building an image–text AI training dataset for a machine translation company’s multimodal image translator. The project runs from February to July 2026.

An image translator has to read the text inside an image and translate it in context. Training one takes large volumes of paired data: the image, the text in it and its translation. Twigfarm produces this data with the experience it gained localizing webtoons and image content, using the LETR AI image editor.

Since 2025 the LETR AI image project has supported JSON uploads of image sentences and index matching, and is used for multimodal dataset work.