A developer operating under the alias hashfunction has published a dataset containing metadata for approximately 5.6 billion public TikTok videos on the Hugging Face platform. The collection, released via the datasocial account, covers the period from July 2014 through October 2026 and is available for free download in Parquet format, totaling 460 GB.
The dataset includes captions, hashtags, sound IDs, on-screen text, and engagement metrics such as views, likes, and shares. It also contains TikTok-specific labels, including AI-generation flags and For You page restrictions. According to the project's documentation, the data was gathered by accessing TikTok's private mobile API using reverse-engineered request signatures and spoofed device identities.
The collection process reportedly harvested 3.23 billion creator profiles, 5.94 billion videos, and 2.8 billion comments over a three-week period without requiring account logins. The data is currently licensed under CC BY-NC 4.0 for research and personal use. While the dataset is free, the developer offers the scraping source code for $1,699 and provides commercial access via a separate website.
This release follows the existence of another 4.5-billion-video set already hosted on Hugging Face, also reportedly sourced from TikTok's mobile API. Official access to TikTok data remains restricted to approved researchers at academic institutions in specific regions like the US, UK, and EU, requiring a formal application process.
