Run anime image classification models on mobile and desktop!
Take a look at the latest release!
- 🗂️ Model manager - select from a range of models
- 🏷️ Tag any image (anime, manga, real photos) with booru tags
- 🔍 Discover what anime characters are in an image
- 📝 Modify the tags output format (spaces, underscores, prompt)
- ⭐ Leveling system based on image rankings
- 🔞 Detects if an image is explicit, questionable, sensitive, or safe
- 🔗 Direct link to tag information
- 🎨 UI theme colors based on input image
- 🌐 Language support: English, Chinese, Japanese, Korean, French, German, Spanish
Models are available in QUINT8 and FP16. View the model catalog for more details.
More models are planned!
WaifuTagger uses modified ONNX models to make mobile inference easier.
Feel free to send a PR to add a new model!
- Input: Uint8Array of rgb data (uint8[1,height,width,3])
- Ouput: Float32Array with sigmoid applied
- Quantization: At least FP16
- File name: "{model-name}-{quant}-mobile.onnx"
For reference, checkout the WaifuTagger Model Collection. Most model repos in the collection should contain a python notebook.