A GUI tool for labeling image from your clipboard or file system using wd tagger

How to install and run

Use the pre-built executable

1. Download the latest release from here
2. Right click on the downloaded file and select Run or Open
3. Wait for the application to start

(Windows excutable gives false positive on some antivirus software, you can build the executable yourself if you don't trust the pre-built one)

Use the source code

1. Create a virtual environment and activate it

python3 -m venv venv

for windows

./venv/scripts/activate

for macos and linux

source venv/bin/activate

2. Install the requirements

pip install -r requirements.txt

3. Run the application

python gui.py

It may take a while to initialize the application for the first time.

How to use

1. Copy an image to your clipboard or select a file
2. Click on the load image from clipboard or load image from file button
3. Click on the analyze image button or press a keybinding.
4. The tags will be displayed in a new window (First time will take a while to download pre-trained model)
5. You can copy the tags to your clipboard by clicking on the copy tags to clipboard button Extra:

• Check Unload model after every analysis can save you some memory, but it will take longer to analyze the image
• You can choose tag format, currently support Booru and Stable Diffusion format

Configuration

After the first run, a config.ini file will be created in the same directory as the script. You can change the configuration there.

[GUI]
shortcut = Ctrl+Shift+I
unload_model_when_done = False
tag_format = booru

[Tagger]
model = wd-swinv2-v3
threshold = 0.35

Default model is wd-swinv2-v3 and I also recommend these models:

• wd-swinv2-v3 (default, with overall good performance)
• wd-convnext-v3 (might deals rotated images better than other models)
• wd-vit-v3 (good at character recognition)
• wd14-moat-v2 (Incase you want to use the old model)

Default confidence threshold is 0.35, lower it if you want more tags (less accurate).

Known issues

• User from china mainland might have trouble downloading the model from huggingface
• macOS keybinding works by excute the script in IDEs (e.g. PyCharm or VSCode), but not in terminal. And it needs you to trust the IDE in System Preferences -> Security & Privacy -> Privacy -> Input Monitoring (Not a safe practice, use at your own risk)
• switch keybinding through GUI crahes on macOS (not sure why)

Copyright

Original code by https://github.com/picobyte/stable-diffusion-webui-wd14-tagger

Public domain, except borrowed parts (e.g. dbimutils.py)