AI Watchdog: OpenVid-1M

Explore original journalism about this data set through AI Watchdog, The Atlantic’s ongoing investigation into the generative-AI industry.

---

OpenVid-1M is a collection of more than 1 million video clips paired with text captions. The videos are taken from YouTube, Pixabay, Pexels, and other sources. The data set was compiled by researchers at ByteDance (TikTok’s parent company) and released in 2024. Researchers gathered videos from multiple existing data sets, including CelebV-HQ and Panda-70M, and filtered and re-captioned the clips, or added captions where they were missing. The data set is hosted on Hugging Face, an AI-development hub. It has been used in experimental contexts, at least, by Amazon, Microsoft, Nvidia, ByteDance, Kuaishou, and others, and has been downloaded more than 380,000 times.