LAION-Audio-XL and CLAP2
LAION-5B and OpenCLIP gave vision open web-scale data and reproducible scaling laws. LAION-Audio-XL and CLAP2 bring the same to audio. We release LAION-Audio-XL, a web-scale audio-text corpus, togethe…
LAION is a non-profit research network building open datasets, models, and tools for reproducible machine-learning research and public education.
Donate
Help us build open pipelines for mathematics and physics, biology, and multimodal world models, and power two lighthouse projects: Modelympics and Project Alexandria.
What we build
Datasets, models, and tools released in the open, so anyone can inspect, reproduce, and build on them.
365 million audio clips with 707 million captions, and CLAP2: open audio–text models that beat every earlier CLAP model.
A 10-million-hour open video dataset for research on video-language, audio-language, and image–text models.
The updated LAION-5B: 5.5 billion text–image-link pairs with additional safety fixes.
Large-scale open implementations and models for contrastive language–image training.
Open language-model baselines and intermediate checkpoints for comparing training datasets.
An open conversational AI assistant designed to make education more accessible.
Why open
LAION exists to broaden access to large-scale machine-learning research. We make reusable resources public so more people can inspect, improve, and learn from them.
Large-scale multimodal datasets, each released with documentation of its access conditions and intended uses.
Weights, training code, and evaluation tools published alongside results, so experiments can be reproduced and compared.
Researchers, engineers, and contributors from around the world, collaborating in the open.
From the network
LAION-5B and OpenCLIP gave vision open web-scale data and reproducible scaling laws. LAION-Audio-XL and CLAP2 bring the same to audio. We release LAION-Audio-XL, a web-scale audio-text corpus, togethe…