Turn images into booru tags inside ComfyUI

What it does

This ComfyUI custom node runs WD14 models to interrogate images and output booru-style tags. It auto-downloads the ONNX models on first use and supports batch processing. You can wire it into a workflow or right-click any image node for a quick one-off interrogation.

The interesting bit

The node blurs the line between pipeline and utility: you can wire it into a formal batch workflow, or just right-click any image preview to pull tags on a whim. It also runs comfortably on CPU via onnxruntime, so you do not need to burn GPU cycles on tagging.

Key highlights

• Auto-downloads WD14 models (MOAT, ConvNextV2, etc.) on first run if missing
• Supports batched inputs for processing multiple images at once
• Right-click any LoadImage, SaveImage, or PreviewImage node for quick interrogation without rewiring your graph
• Runs fully offline after initial model caching
• CPU-friendly via onnxruntime; GPU support is available but explicitly unsupported

Caveats

• The author states they cannot provide support for onnxruntime-gpu, so CPU inference is the practical default
• Manual offline setup requires downloading specific .onnx and .csv files with exact filenames

Verdict

Worth adding if your ComfyUI workflows already involve image tagging and you want interrogation without leaving the node graph. If booru-style metadata is irrelevant to your work, this node will not change your mind.