From the model card
Fields below are copied from the tags and counters on the HuggingFace repository SmilingWolf/wd-swinv2-tagger-v3 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- SmilingWolf
- Library
- timm
- Weight formats
- ONNX, safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Downloads (HF counter at last fetch)
- 414,764
- Likes (HF counter at last fetch)
- 95
- Model card
- https://huggingface.co/SmilingWolf/wd-swinv2-tagger-v3
Use cases
- Auto-tagging anime illustration datasets for training diffusion models
- Filtering image collections by character, style, or content rating
- Building image search indices for anime art databases
- Generating descriptive tags for text-to-image prompt engineering
- Content moderation labelling for anime platforms
Pros
- SwinV2 backbone provides better spatial hierarchy than ViT for fine-grained tag prediction
- V3 covers more tags than V2 with improved precision on rare character attributes
- 86 likes with heavy use in Stable Diffusion dataset pipelines
- timm-compatible; easy to integrate into existing vision pipelines
Cons
- Danbooru-trained; generalises poorly to non-anime imagery
- Rating tags (explicit/questionable) require careful handling in production systems
- No explicit license; verify terms before commercial dataset use
- SwinV2 inference is slower than ViT-based taggers for large batches
Tags
timmonnxsafetensorslicense:apache-2.0region:us