A ComfyUI extension allowing the interrogation of booru tags from images.

Based on SmilingWolf/wd-v1-4-tags and toriato/stable-diffusion-webui-wd14-tagger
All models created by SmilingWolf

1. git clone https://github.com/pythongosssss/ComfyUI-WD14-Tagger into the custom_nodes folder
   • e.g. custom_nodes\ComfyUI-WD14-Tagger
2. Open a Command Prompt/Terminal/etc
3. Change to the custom_nodes\ComfyUI-WD14-Tagger folder you just created
   • e.g. cd C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-WD14-Tagger or wherever you have it installed
4. Install python packages
   • Windows Standalone installation (embedded python):
     ../../../python_embeded/python.exe -s -m pip install -r requirements.txt
   • Manual/non-Windows installation
     pip install -r requirements.txt

Add the node via image -> WD14Tagger|pysssss

Models are automatically downloaded at runtime if missing.

Supports tagging and outputting multiple batched inputs.

• model: The interrogation model to use. You can try them out here WaifuDiffusion v1.4 Tags. The newest model (as of writing) is MOAT and the most popular is ConvNextV2.
• threshold: The score for the tag to be considered valid
• character_threshold: The score for the character tag to be considered valid
• exclude_tags A comma separated list of tags that should not be included in the results

Quick interrogation of images is also available on any node that is displaying an image, e.g. a LoadImage, SaveImage, PreviewImage node.
Simply right click on the node (or if displaying multiple images, on the image you want to interrogate) and select WD14 Tagger from the menu

Settings used for this are in the settings section of pysssss.json.

Simplest way is to use it online, interrogate an image, and the model will be downloaded and cached, however if you want to manually download the models:

• Create a models folder (in same folder as the wd14tagger.py)
• Use URLs for models from the list in pysssss.json
• Download model.onnx and name it with the model name e.g. wd-v1-4-convnext-tagger-v2.onnx
• Download selected_tags.csv and name it with the model name e.g. wd-v1-4-convnext-tagger-v2.csv

onnxruntime (recommended, interrogation is still fast on CPU, included in requirements.txt)
or onnxruntime-gpu (allows use of GPU, many people have issues with this, if you try I can't provide support for this)

• 2023-05-14 - Moved to own repo, add downloading models, support multiple inputs