A dataset of about 5.8 billion image-text pairs, filtered with CLIP.
LAION-5B is a giant fridge for drawing AI. It is covered with billions of photos and tiny sticky notes.
CLIP checked every pair first. It helps train text-to-image models. It also shows up in copyright fights and data-cleaning arguments.
CLIP
CLIP often removes image-text pairs with a bad match.
Stable Diffusion
Early Stable Diffusion training used a subset of LAION-5B.
Text-to-Image Generation
LAION-5B gives these models huge piles of matched images and text.
Copyright
Its web images keep it in copyright fights.