LAION

LAION, short for Large-scale Artificial Intelligence Open Network, is a non-profit association registered in Germany whose mission is to keep large-scale AI research accessible to everyone. Rather than training a single flagship model, LAION produces public infrastructure: multi-billion-item multimodal datasets, open model weights, reproducible training pipelines and evaluation tooling that academic groups, independent researchers and startups can all build on.

Its catalog spans several well-known projects. Re-LAION-5B is the updated version of the original LAION-5B text–image dataset, containing roughly 5.5 billion image–text link pairs with additional safety and filtering fixes. LAION-BVD (Big Video Dataset) is a 10-million-hour open video dataset aimed at video-language and audio-language research. OpenCLIP provides scalable open implementations and pretrained models for contrastive language–image training, while Open-sci-ref supplies open language-model baselines for comparing training datasets. The BUD-E project explores an open conversational AI assistant for education.

Everything LAION releases is free to use and documented with access conditions and intended uses. The organization is funded through donations and compute contributions and collaborates with a global community of researchers, engineers and volunteer contributors. Users typically download datasets from Hugging Face, read code on GitHub and follow release notes on the LAION blog, which documents ongoing work such as the Modelympics and Project Alexandria lighthouse initiatives.