LAION

Data for multimodal research

LAION organizes public research resources for training and studying machine-learning systems. Its work includes datasets, models and code, with an emphasis on making resources available to researchers outside large commercial labs. [1]

The LAION-5B release supplied metadata for 5.85 billion image-text pairs filtered with CLIP. URLs, captions and associated representations let researchers select and retrieve material for multimodal experiments; the release is not a transfer of ownership of billions of images. [2]

The original announcement licenses the metadata under CC BY 4.0 and explicitly says the images retain their own copyright. That distinction matters when using LAION as a training-data source: access to an index and permission to use every indexed image are separate questions. InfoQ's later report corroborates the release, while the primary announcement establishes its March 31 date. [2][3]