Joytag is a Danbooru style model that has been trained on the Danbooru dataset in addition to photographs and "other content not well represented in the main dataset". Basically it generalizes better to non-anime/manga images as a result of this training, while maintaining danbooru style tagging and a small model footprint. I think it would act as a good supplement to the other danbooru style taggers already in taggui (the WD series).
Joytag is a Danbooru style model that has been trained on the Danbooru dataset in addition to photographs and "other content not well represented in the main dataset". Basically it generalizes better to non-anime/manga images as a result of this training, while maintaining danbooru style tagging and a small model footprint. I think it would act as a good supplement to the other danbooru style taggers already in taggui (the WD series).