Built an open-source Qwen3-VL auto-tagging toolkit: ComfyUI node + standalone CLI (XMP + JSONL)

[ekkonwork]

Hi everyone,

I built an open-source auto-tagging toolkit around Qwen3-VL for stock-style image metadata generation.

What it does:

• Generates a title and keyword list from an image
• Saves metadata to JSONL
• Optionally writes XMP fields directly into output files (Title / Description / Subject)

Two versions:

• ComfyUI node: https://github.com/ekkonwork/comfyui-qwen3-autotagger
• Standalone CLI: https://github.com/ekkonwork/qwen3-vl-autotagger-cli

CLI quick example:
python -m qwen3_vl_autotagger_cli.cli "./images" --recursive --output-dir "./outputs" --write-xmp

Notes:

• Default model: Qwen/Qwen3-VL-8B-Instruct
• Tested on Colab T4
• CLI repo includes a production Colab runner

If useful, I can also share a short post about prompt tuning for more stock-friendly keyword quality.

Thanks.