What it does

This extension plugs into Automatic1111’s Web UI to interrogate images and spit out booru-style tags — the kind used on anime image boards. Feed it a single image or a batch, and it returns descriptive labels using models like DeepDanbooru or the Waifu Diffusion 1.4 tagger. The WD14 model downloads itself from HuggingFace on first run; DeepDanbooru models need manual placement in a models/deepdanbooru folder tree.

The interesting bit

The README opens with a blunt “This branch is unmaintained!” and immediately redirects to a fork by picobyte. That’s refreshingly honest in a space where dead projects often just rot. The code itself is largely borrowed — the author admits most of it came from DeepDanbooru and MrSmilingWolf’s tagger — so this is more plumbing than invention.

Key highlights

• Supports multiple interrogator models: WD14 (auto-download), DeepDanbooru variants, and an e621 model
• Batch processing for multiple images
• Korean translation of the README exists
• Public domain licensing (except borrowed utilities like dbimutils.py)
• Includes a model comparison doc for choosing between taggers

Caveats

• This branch is explicitly unmaintained; the README points to picobyte’s fork as the successor
• Installation requires a full Stable Diffusion restart, not just a refresh
• DeepDanbooru setup involves manual folder wrangling with a specific nested structure

Verdict

Worth a look if you’re still running Automatic1111 and need to tag training data or organize image collections. Skip it if you’re on ComfyUI or expect active support — this repo is essentially a redirect sign with 1,417 stars.