AutoTag

AutoTag is a feature where a local AI analyzes images and suggests tags. It’s optional, and everything works fine with manual tags alone.

What you can do

• Imported images get tags automatically, reducing tagging work
• Filter and browse by AutoTags (the “AutoTags” in the sidebar)
• Similar-image search via “Find similar” becomes available

Set it up

Install the model from the initial setup, or from the Desktop settings, with “Install model”. It downloads the open-source JoyTag model and runs it on your PC.

All analysis happens locally. Your images are never sent externally or used to train AI.

How easily tags appear

In settings, “How easily tags appear” controls how many tag suggestions are picked up. Raise it for more suggestions, lower it for only high-confidence ones. Usually there’s no need to change it.

Connect to manual tags with mapping

AutoTag keys are in English (for example, red_hair). Map them to your own tags, and they work just like your usual tags in search and filtering. Even without mapping, you can use AutoTags on their own for filtering.

Set this up under Settings > Auto-Tag in the sidebar.

Review AutoTags

AutoTag Statistics lists the detected AutoTags. Search them, and sort by prediction count. Select an AutoTag to see what it’s mapped to (“No mapping configured” if it isn’t) and the stacks with that AutoTag.

Create a mapping

1. Select an AutoTag and press Create Mapping.
2. Under Map to Tag, pick an existing tag, or type to create a new one.

Edit or delete a mapping

For AutoTags that are already mapped, edit or delete the mapping from where the mapped tag is shown.

If you enable it later

For stacks you’ve already imported, apply AutoTags with “Refresh” per stack, or with Full Refresh per library.

System requirements

It runs on CPU alone. On Windows with an NVIDIA GPU, enabling “Generate tags with GPU” in the AutoTag settings speeds up analysis.

• This option is only selectable when an NVIDIA GPU is detected. If you can’t select it, update your NVIDIA driver first and press “Check AutoTag”
• You don’t need to install the CUDA Toolkit manually

If it doesn’t seem to be working, check the JoyTag Server status (Running / Not Available). While it’s Not Available, no tags are generated and no mapping candidates are collected.