Multilingual text-to-image model makes AI more inclusive

Through SURF and NWO, PhD student Dheeraj Varghese from the University of Amsterdam and his colleague Mohammad Derakhshani gained access to the LUMI supercomputer to develop a text-to-image model called NeoBabel. Their aim was for this model to be able to generate images based on prompts in no fewer than six very different languages: English, Dutch, Chinese, Hindi, French and Persian.

The massive computing power of LUMI enabled the team to expand their training dataset from 40 to 124 million image-label pairs and train the model at a scale not possible with Dutch computing resources alone. NeoBabel has been a success, and the result can be viewed online by anyone. Additionally, all of NeoBabel’s material is available as open source, and anyone is free to use it and build on it to make AI more inclusive, across linguistic and cultural boundaries.

Read the full original use case on SURF’s website.