What it does

It’s a web UI for assembling Stable Diffusion prompts from Danbooru/NovelAI tags without memorizing thousands of obscure strings. You search, click, adjust weights, and export. It also catalogs Textual Inversion embeddings and Hypernetworks with download links, since keeping track of those is its own part-time job.

The interesting bit

The whole taxonomy lives as YAML in data/tags/**/*.yaml — community-editable, version-controlled, and surprisingly disciplined. There’s even npm run dupcheck because tag collisions are apparently a real problem at scale. The project also handles the weird A1111 syntax like prompt editing and alternating words, which most builders ignore.

Key highlights

• Bilingual tag names (English/Chinese) with optional images, descriptions, and Danbooru Wiki links
• Weight tuning and advanced prompt syntax support (editing, alternating words)
• Import and auto-parse existing tag blobs back into the builder
• Preset combinations with previews for common character/style recipes
• Embedding and Hypernetwork indexing with SHA256 verification and HuggingFace hosting for large .pt files
• CLI helpers for image cropping, compression, and embedding format conversion

Caveats

• Building requires Font Awesome v6 Pro (paid license) due to some icon usage; Free icons work for dev but aren’t the default
• Masonry layout and Collapse UI elements currently don’t play nice together (listed in TODO)
• Embedding uploads only accept PNG-chunk format, not raw .pt files — there’s a Colab notebook for conversion, but it’s an extra hop

Verdict

Worth a bookmark if you generate anime-style images and spend more time guessing tags than iterating. Skip it if you’re already happy with your prompt workflow or don’t touch the Danbooru tag ecosystem.