LAION releases open dataset of 80 million videos, 300 million frames...

LAION releases open dataset of 80 million videos, 300 million frames, and audio tracks

The German non-profit group LAION has released the open video dataset LAION-BVD — 80 million downloaded videos totaling 10 million hours, cut into 55 million clips with synthetic descriptions, plus 300 million frames and two audio collections of 1.7 million and 10 million tracks.

Compiled by Sergey KostenchukPublished 2026-08-31Updated 2026-09-18

2026-08-31 Research HuggingFace

🤖 LAION releases open dataset of 80 million videos

The non-profit group LAION has released the video dataset LAION-BVD: 80 million videos totaling 10 million hours were collected from 1.3 billion CommonCrawl links, cut into 55 million clips with synthetic descriptions, plus 300 million frames and audio of 1.7 million and 10 million tracks. According to arXiv:2608.24845, ViCLIP and CLAP on BVD outperform models on InternVid by up to +2.1%.

🌍 The largest video datasets for multimodal pre-training are concentrated at proprietary companies. BVD for the first time opened 10 million hours of video to the scientific community in three modalities — video-text, audio-text, image-text.

👤 You can fine-tune your video or audio model now: URL versions of all sub-samples are freely available on Hugging Face. The license is for research, commercial use is prohibited, and the full corpus with media files is provided upon application.

Source 1: https://arxiv.org/abs/2608.24845

Source 2: https://projects.laion.ai/bvd/index.html

Sources