Danbooru2020 is a large-scale anime image database containing more than 4.2 million images with more than 130 million text annotations that describe the image content in detail. This dataset can be used for machine learning, such as image recognition and generation. This dataset has been used for various applications, especially generative modeling. Danbooru20xx is improved and updated every year based on images and metadata from previous years. Previous versions are: Danbooru2017, Danbooru2018, Danbooru2019.