This public research dataset contains compact NPZ v2 samples inside uncompressed
WebDataset tar shards. It has 29510 prepared samples from
8698 source videos; 57 sources were
recorded as terminal skips.
The real and fake values are trusted scraper assumptions, not independently
verified ground truth. Pre-2021 videos are proxy-real and creator-disclosed altered
videos are proxy-fake. Do not report model agreement as… See the full description on the dataset page:
https://huggingface.co/datasets/gonnerthetooner/orislop-youtube-prepared-v2.
