Laion

LAION is a nonprofit open-AI organization that makes machine-learning resources available for public research and education. Its project library includes large datasets, open-source models, and utilities for working with multimodal data. The dataset collection covers image-text resources such as LAION-400M and LAION-5B, a multilingual corpus of 5.85 billion CLIP-filtered image-text pairs. Other releases include generated image captions, translated samples, high-resolution image subsets, aesthetic-image subsets, watermark-detection data, and audio datasets with captions, labels, and metadata. LAION also lists projects spanning 3D, video, and other modalities. Its model projects include OpenCLIP, CLAP for audio-text representation, Multilingual-CLIP, NSFW Detection, ClipCap, and a DALL-E 2 implementation. For practitioners building datasets or retrieval systems, tools such as img2dataset can turn large collections of image URLs into packaged datasets, while Clip Retrieval supports computing CLIP embeddings and building image-text retrieval systems. Researchers, educators, and open-source machine-learning developers can use these resources for model training, benchmarking, exploration, and dataset construction.