Danbooru2021 mirror/dataset

Posted under General

The fifth edition of my Danbooru mirror/dataset, Danbooru2021, is now live:


This updates my previous Danbooru2020 (topic #17915) through 31 December 2021, adding an additional +1.1TB of images (+725k) and +32m tags.

It can be downloaded via rsync; the BitTorrent option has been removed.

Major changes: no more BitTorrent; and the metadata JSON export now uses the new `danbooru1` BigQuery mirror, which is much more comprehensive and includes other kinds of metadata beyond the post metadata in the old BQ mirror.

An example of Danbooru2020 use from last year is This Anime Does Not Exist: https://thisanimedoesnotexist.ai/