Run SmilingWolfs WD Convnext Tagger v3 model on mobile and WD SwinV2 Tagger v3 on desktop!
An upcoming version for WaifuTagger! This update will be released the week of 08/09 (delayed).
Version 2.1 is almost an entire rewrite. Here are a few highlights:
- Supports multiple models: WD taggers, Camie v2, and OppaiOracle v1.1
- Desktop has been migrated to Rust (Tauri v2). ONNX is now running natively instead of through WASM.
- In-app updates. Always know when an update is available!
- Improved dynamic theming.
- App translations: English, Japanese, Chinese, Korean, French, German, Spanish.
- Many bug fixes and UI tweaks.
Android and Desktop (Windows, MacOS, Linux)
Take a look at the latest release!
To run the model on mobile, I quantized it and added preprocessing using onnxruntime-extentions. This model is available on my HuggingFace. You can also find more details about the process there.
Desktop is using p1atdev's transformers version of the WD SwinV2 Tagger v3 model due to some ops (decodeImage) not working with onnxruntime-web.
The WASM backend is used for inference.
If you get bad results, it's highly likely that I did something wrong with the preprocessing. Any tips for improvement would be appreciated.
- This app uses expo-share-intent and will need additional setup for iOS.