DeepDanbooru is an open-source, TensorFlow-based image tagging system that automatically assigns Danbooru-style tags to anime and illustration artwork. Trained on millions of tagged illustrations, its multi-label classifier can recognize characters’ visual attributes — hair and eye color, clothing, accessories, poses and more — and output them as searchable tags. The project has around 3,000 stars on GitHub and continues to receive updates.
Booru-style image boards live and die by their tags, but manual tagging does not scale to large collections. DeepDanbooru automates it: feed in an illustration and the model quickly returns hundreds of possible tag predictions with confidence scores, which you can threshold into a final tag set. It ships as a command-line tool with a pretrained model, so it can be dropped into collection-management scripts and dataset-building pipelines. The tagger has also become a standard component in the anime AI ecosystem — dataset curators for image-generation models use it to auto-label training data, and several popular WebUI extensions integrate it directly. Evaluators and rating-based filtering are included as well, so low-confidence tags can be pruned automatically to keep large auto-tagged collections clean and searchable.
DeepDanbooru is free and open source under the MIT license, including its pretrained models. There is no hosted version or paid tier — everything runs locally on your machine.
DeepDanbooru is for illustration archivists, dataset builders and AI art practitioners who need fast, automatic tagging of anime-style images for search, curation or model training.