---
title: "Metadata of 4.5 billion TikTok videos, allegedly on Hugging Face"
description: "One X post claims a 289 GB metadata dataset of TikTok videos and lists the virality questions it could answer."
date: 2026-09-07T12:00:00+02:00
lastmod: 2026-11-03T00:46:22+01:00
url: https://klimek.link/posts/2026-09-07-tiktok-metadata-dataset/
author: "Marcin Klimek"
source: https://klimek.link/posts/2026-09-07-tiktok-metadata-dataset/
---

# Metadata of 4.5 billion TikTok videos, allegedly on Hugging Face

Source: [@rpl_0x on X](https://x.com/rpl_0x/status/2096925940051853489) (in Polish)

The claim, in a Polish post on X: someone uploaded metadata for about 4.5 billion TikTok videos to Hugging Face, with captions, views, likes, saves, shares, length, country, sound and an ad flag for every clip, and no video. The post lists what you could study with it, from save rates as a predictor of virality to the hooks in top captions and the difference between ads and organic posts. It doesn't name the uploader or say how the data was gathered.

**TL;DR**

- Source: one Polish post by @rpl_0x on X (249 likes, 14 reposts when saved)
- Scope: metadata of ~4.5 billion TikTok videos, ~289 GB, no video files
- Columns: caption, views, likes, comments, shares, saves, length, date, country, language, sound ID, ad flag
- Questions: best video length, top caption phrases and hooks, above-average sounds
- Also: do save and share rates predict viral videos, trends across countries, ads vs organic
- Hunt for odd videos with huge saves or shares relative to views
- Method idea: SQL or DuckDB plus LLM sorting into storytelling, education, humor, ragebait, tutorials
- No uploader or collection method named; dataset not verified here

Source: https://klimek.link/posts/2026-09-07-tiktok-metadata-dataset/