LAION-5B
In plain wordsA free list of nearly six billion links to online images paired with their captions, released in 2022 and used to train image generators such as Stable Diffusion.
An open dataset of 5.85 billion image–text pairs gathered from Common CrawlCommon Crawl, prop. n.: A nonprofit's free archive of the public web, begun in 2008 and now refreshed with snapshots of billions of pages every month or two. Most large language models learned much of what they know from text taken from it. by LAION (Large-scale Artificial Intelligence Open Network), a German nonprofit; 2.32 billion have English captions and the rest are in over 100 other languages or of undetected language. Pairs were found by parsing web pages for images with alt text and kept only if CLIPCLIP, prop. n.: An OpenAI model that learned to match pictures with their captions from hundreds of millions of web examples, so it can judge how well any piece of text describes any image. judged image and caption sufficiently similar. LAION distributed URLs, captions, CLIP embeddingsembedding, n.: A list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance. and safety and watermark scores rather than the images themselves. Subsets such as LAION-2B-en and LAION-Aesthetics trained Stable DiffusionStable Diffusion, prop. n.: A widely used image-generating AI whose model files were released for anyone to download, so people could run it on their own computers and adapt it. and open CLIP replications, making it the foundation of open text-to-imagetext-to-image model, n.: An AI system that creates a picture from a written description, such as 'a fox reading a newspaper, in watercolor', usually within seconds. research. Its scale without manual review exposed problems — copyrighted works, private and medical photos and, in a December 2023 Stanford Internet Observatory report, links to child sexual abuse material — prompting LAION to withdraw it and release a cleaned Re-LAION-5B in August 2024.
We filtered LAION-2B-en to images above 512 pixels with an aesthetic score over 5.
Origin
Assembled by a volunteer community around Christoph Schuhmann, who had released the smaller LAION-400M in 2021; published in 2022 and described in 'LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models' (Schuhmann, Romain Beaumont, Richard Vencu and colleagues), NeurIPS 2022 Datasets and Benchmarks Track.
Usage note
Re-LAION-5B removed 2,236 links matched against hash lists from child-protection organizations, leaving about 5.5 billion pairs. In Kneschke v. LAION the Hamburg Regional Court held in September 2024 that downloading images to build the dataset fell under Germany's text-and-data-mining exception for scientific research; the Hanseatic Higher Regional Court dismissed the photographer's appeal in December 2025.
Related
- CLIPCLIP, prop. n.: An OpenAI model that learned to match pictures with their captions from hundreds of millions of web examples, so it can judge how well any piece of text describes any image.
- Stable DiffusionStable Diffusion, prop. n.: A widely used image-generating AI whose model files were released for anyone to download, so people could run it on their own computers and adapt it.
- text-to-image modeltext-to-image model, n.: An AI system that creates a picture from a written description, such as 'a fox reading a newspaper, in watercolor', usually within seconds.
- Common CrawlCommon Crawl, prop. n.: A nonprofit's free archive of the public web, begun in 2008 and now refreshed with snapshots of billions of pages every month or two. Most large language models learned much of what they know from text taken from it.
- web scrapingweb scraping, n.: Using software to automatically copy information from websites — text, prices, images — in bulk, which is how much of the data used to train AI was gathered.
- data provenancedata provenance, n.: The documented history of a piece of data: where it came from, who created and changed it, under what permissions, and how it was processed before reaching an AI system.
Sources
- Schuhmann, C. et al. (2022). LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models. NeurIPS 2022 Datasets and Benchmarks Track. arXiv:2210.08402. arxiv.org/abs/2210.08402
- LAION (2024). Re-LAION-5B release announcement. LAION blog, 30 August 2024. laion.ai/blog/relaion-5b
Mentioned in
Stable DiffusionStable Diffusion, prop. n.: A widely used image-generating AI whose model files were released for anyone to download, so people could run it on their own computers and adapt it., CLIPCLIP, prop. n.: An OpenAI model that learned to match pictures with their captions from hundreds of millions of web examples, so it can judge how well any piece of text describes any image., LAIONLAION, prop. n.: A mostly volunteer group, registered as a German nonprofit, that compiled free lists of billions of online pictures and their captions. Stable Diffusion and other image generators were trained on them., web scrapingweb scraping, n.: Using software to automatically copy information from websites — text, prices, images — in bulk, which is how much of the data used to train AI was gathered.