Danbooru2020

A large-scale anime image database with 4.2m+ images annotated with 130m+ text tags describing image contents in detail; it can be useful for machine learning purposes such as image recognition and generation. It has been applied to a wide variety of applications, particularly generative modeling.

Danbooru20xx is updated annually with the previous years' images & metadata improvements. Previous iterations: Danbooru2017, Danbooru2018, Danbooru2019.

Source: Danbooru2020.