Laion
Unlock AI potential: vast datasets, cutting-edge models, free access, eco-conscious.. [Free]
Last verified:
What is Laion?
LAION (Large-scale Artificial Intelligence Open Network) is a German non-profit organization that provides open-source datasets, tools, and machine learning models to democratize ML research for the general public. The organization is best known for releasing massive image-text pair datasets like LAION-400M (400 million English image-text pairs) and LAION-5B (5.85 billion multilingual CLIP-filtered image-text pairs), which have been used to train high-profile text-to-image models including Stable Diffusion and Imagen.
Key features include open datasets (LAION-400M, LAION-5B, LAION-Aesthetics, LAION5B High-Res, Laion-COCO), released models (OpenCLIP, CLIPCap, NSFW Detection, CLAP), and tools (img2dataset for dataset reconstruction, Clip Retrieval for image search, Clip Benchmark). LAION-Aesthetics provides subsets filtered by aesthetic scoring models, while LAION-5B High-Res contains images with resolution >= 1024.
LAION is designed for machine learning researchers, data scientists, AI developers, educators, and anyone interested in large-scale ML research. The organization teaches basics of large-scale ML research and data management, promotes efficient use of energy and computing resources to address climate change, and encourages open public education through reusable models and datasets that don't require training from scratch.
Laion pricing
Pricing model: Freemium
100% FREE - LAION is a 100% non-profit organization providing all datasets, tools, and models completely free to the general public. The organization is funded by donations and public research grants. There are no paid plans, subscription tiers, or premium features. All released datasets (LAION-400M, LAION-5B, LAION-Aesthetics, LAION5B High-Res, Laion-COCO), models (OpenCLIP, CLIPCap, NSFW Detection), and tools (img2dataset, Clip Retrieval, Clip Benchmark) are available at no cost.
Laion pros
- 100% free to use with no paid plans
- 100% non-profit organization focused on public good
- Truly open AI with no restrictions
- Massive LAION-5B dataset with 5.85 billion image-text pairs
- Multilingual support in LAION-5B dataset
- Datasets used to train Stable Diffusion and Imagen
- OpenCLIP model released for public use
- img2dataset tool for reconstructing datasets
- Clip Retrieval tool for image search capabilities
- LAION-Aesthetics subset for high-quality images
- LAION5B High-Res subset with images >= 1024 resolution
- Promotes efficient energy and computing resource use
- Funded by donations and public research grants
- Members from all over the world
- Teaches large-scale ML research basics
- NSFW Detection model released
- CLAP audio/text contrastive model in development
Laion cons
- Datasets are only URL indexes, not actual images
- Users must reconstruct images using img2dataset tool
- No official hosting of reconstructed image datasets
- Links can lead to disturbing or discomforting images
- Copyright concerns with scraped web content
- Cannot remove data from past releases shared via torrents
- 일부 ongoing projects still in planning or started status
- No commercial support or SLA guarantees
Frequently asked questions about Laion
Does LAION datasets respect copyright laws?
LAION datasets are simply indexes to the internet - lists of URLs to original images together with ALT texts found linked to those images. While LAION downloaded and calculated CLIP embeddings of pictures to compute similarity scores, they subsequently discarded all photos. Any researcher using the datasets must reconstruct the image data by downloading the subset they are interested in using the img2dataset tool.
What should I do if links in the dataset show my copyrighted data?
LAION is a non-profit research organization studying how learning algorithms work. TDM exemptions are valid for LAION according to Art. 3 EU TDM exemption and §60d UrhG of German law for research. LAION is permitted to use copyrighted material as data for conducting research on learning algorithms and foundation models as training outcome.
Do the datasets contain disturbing images?
No, but links in the datasets can lead to images that are disturbing or discomforting depending on the filter or search method employed. The datasets themselves are just URL indexes with ALT text.
What if I found a dataset containing images while searching on the internet?
Any dataset containing images is not released by LAION - it must have been reconstructed with the provided tools by other people. LAION does not host and does not provide links on their website to access such reconstructed datasets. Users should refer only to links provided for official released data.
How can I protect my personal data under GDPR if I found my name and picture in the dataset?
If you found your name only on the ALT text data and the corresponding picture does NOT contain your image, this is not considered personal data under GDPR terms. Your name associated with other identifiable data is. If the URL or the picture has your image, you may request a takedown of the dataset entry in the GDPR page. LAION provides a takedown form, and upon submission, they will investigate and if verifiable, remove the entry from all data repositories they control, including current data and future releases.
Do LAION scripts respect robots.txt instructions?
Despite the 'Crawling at Home' project name, LAION is not crawling websites to create datasets. Common Crawl did the crawling part in the past and they did respect the robots.txt instruction. LAION only analyzes their data and then looks at pictures to assess their value concerning the provided alt text.
What is LAION-5B?
LAION-5B is a dataset consisting of 5.85 billion CLIP-filtered image-text pairs, of which 2.32B contain English language. It is multilingual and has been used to train next-generation image-text models including Stable Diffusion and Imagen.
What is LAION-Aesthetics?
LAION-Aesthetics is a subset of LAION-5B filtered by a model trained to score aesthetically pleasing images. There is also LAION-Aesthetics v2 with an updated aesthetic model. These subsets help users find higher quality, more visually appealing image-text pairs.
What tools does LAION provide?
LAION provides several released tools including img2dataset (for reconstructing datasets from URL indexes), Clip Retrieval (for image search and retrieval), Crawlingathome-gpu-hcloud, and Clip Benchmark (currently in progress). These tools help researchers work with LAION datasets effectively.
How is LAION funded?
LAION is funded by donations and public research grants. As a 100% non-profit organization, their aim is to open all cornerstone results from large-scale machine learning to all interested communities without charging for access to their datasets, models, or tools.