Training-image rights and downstream scope for WD SwinV2 Tagger v3
Hello. We are evaluating WD SwinV2 Tagger v3 at revision 627aef95638667ddcaa3ac8ae625e88ea5b02f51 for an offline character-portrait authoring tool. The model repository declares Apache-2.0, while the card says Dataset v3 used Danbooru images through image id 7220105 with the documented modulo split. Could you please identify the source or license basis used for those training images and any restrictions that downstream commercial users of the published weights should account for? In particular, please clarify whether the repository's Apache-2.0 declaration is intended to cover commercial inference and redistribution of the published weight files despite the upstream image corpus, and whether any additional attribution, non-commercial, or redistribution terms apply. A stable model-card, dataset-rights, or license link would be appreciated. We will keep training-image rights recorded as missing until an explicit publisher response and source terms are available. Thank you.