Dataset
LAION
Open image-text dataset (5B pairs) that trained Stable Diffusion and many CLIP models.
Definition
LAION-5B is an open dataset of 5 billion image-URL/alt-text pairs scraped from Common Crawl. It enabled the first open large-scale image-text models â Stable Diffusion, OpenCLIP. The 2023 CSAM controversy led to filtered re-releases (LAION-COCO, Re-LAION).
Common use cases
- Image-gen training
- CLIP training
- Multimodal research