United Statescountry
DevelopingOther

Developer Publishes Metadata for 5.6 Billion TikTok Videos on Hugging Face

A developer named hashfunction uploaded metadata for about 5.6 billion public TikTok videos to Hugging Face, sourced from TikTok's private mobile API without login.

Why it matters

The dataset reaches an area TikTok has kept hard to access, since its terms bar automated scraping without written approval and the developer's own account describes pulling the data from a private mobile API without login. TikTok has not responded in the material available, so whether the scrape breaches its terms is unresolved. The commercial routing to datasocial.ai separates free non-commercial access from paid offerings, including source code at $1,699.

Key facts

A developer going by the handle hashfunction posted metadata for about 5.6 billion public TikTok videos on Hugging Face under the datasocial account, free to download under a non-commercial licence.

5,600,000,000 videosReported by Decrypt

The dataset covers the period from July 2014 through October 2026.

Reported by Decrypt

The dataset is 460 GB of Parquet files, one per month, and holds metadata rather than video footage.

460 GBReported by Decrypt

Each row carries the caption, hashtags, sound ID, on-screen text, TikTok Shop product and seller IDs, and counts for views, likes, comments, shares, saves and downloads.

Reported by Decrypt

Some fields are TikTok's own labels, including whether a video is flagged as AI-generated and whether TikTok keeps it off the For You page, and the AI flag is often empty for older videos drawn from an archive.

Reported by Decrypt

The developer's write-up says the data came from TikTok's private mobile API with no login.

Reported by Decrypt

TikTok's terms bar automated scraping without written approval.

Reported by Decrypt

The dataset card routes commercial use, creator profiles and daily updates to datasocial.ai, which also sells the scraper's source code for $1,699.

1,699 USDReported by Decrypt

What happened

Decrypt reported that a developer using the handle hashfunction uploaded metadata covering roughly 5.6 billion public TikTok videos to the open-source AI repository Hugging Face, published under the datasocial account and free to download under a non-commercial licence. The dataset spans July 2014 through October 2026 and consists of 460 GB of Parquet files, one per month, holding captions, hashtags, sound IDs, on-screen text, TikTok Shop product and seller IDs, and engagement counts for views, likes, comments, shares, saves and downloads, rather than any video footage. Decrypt said the developer's own write-up states the data came from TikTok's private mobile API without login, and that TikTok's terms bar automated scraping without written approval. Some fields carry TikTok's own labels, including whether a video is flagged as AI-generated and whether it is kept off the For You page, with the AI flag often empty for older videos that come from an archive. The dataset is not gated, and Hugging Face's counter showed 1,181 downloads. Decrypt also reported that the dataset card routes commercial use, creator profiles and daily updates to datasocial.ai, which sells the scraper's source code for $1,699. No response from TikTok appears in the available material, and only Decrypt has covered this.

How the story developed

  1. First report

    First report by Decrypt

    Someone Scraped 5.6 Billion TikTok Videos and Put the Data on Hugging Face for Free

    Decrypt

  2. Update

    hashfunction publishes 5.6 billion TikTok video records to Hugging Face

    The upload went live under the datasocial account covering July 2014 through October 2026, with the dataset card pointing commercial use, creator profiles and daily updates to datasocial.ai.

    Decrypt

Affected entities

Original sources

PublisherReportRolePublished
Decrypt
Crypto media
Someone Scraped 5.6 Billion TikTok Videos and Put the Data on Hugging Face for FreeOriginal