GALIAIS-Nodes AI Tag Analyzer

Here's the workflow problem this node solves: you've got a wall of Danbooru tags - 1girl, long hair, red dress, standing, park, sunlight - and you want a structured read on what's actually in there before you send it down the line. GALIAIS-Nodes is built around a 10-layer character prompt system, and the AI Tag Analyzer is the front door for the AI-refinement branch: it breaks your prompt into categories and gives it a quality score, so downstream AI nodes have something structured to work from instead of a raw string.

Two inputs matter here, and neither needs an AI provider. 提示词 is your tag list or prompt, and 允许NSFW (allow NSFW) is a boolean - flip it on if your analysis is allowed to handle adult content, because the model's job here is classification and it needs permission before it labels the spicy tags rather than sidestepping them. There are two optional inputs: DB (a loaded Danbooru database from the pack's DB loader, which lets the analyzer normalize tags against the real dictionary) and 项目配置 (project config). Neither is required to get an answer.

What comes out

Three outputs, and the middle one is the one you'll actually read:

• 分类摘要 (classification summary) - a human-readable breakdown of the prompt by category. This is the "what am I actually describing" view.
• 分析JSON - the structured version, meant to be fed into other nodes like the AI Natural Prompt Writer, which specifically wants Tag分析JSON on its input.
• 质量评分 (quality score) - an integer rating. Don't read it as an absolute law; it's a rough grade for spotting when your prompt is thin or messy, not a verdict on your art.

There's a real debate in the community about how much stock to put in LLM quality scores - they're not the same as a model's own score_9 training tags - so treat this as a rough guide for "is this prompt coherent," not as a number to maximize.

Where it fits

The natural flow in this pack is: build your tag sections in the character nodes → run the whole thing through the Tag Analyzer → hand 分析JSON to the Natural Prompt Writer for a natural-language rewrite, or to the Conflict Resolver when your diagnostics flag trouble. Because the analyzer itself doesn't call an API (it's a rule-based classifier over the tag structure), you can use it before you've even configured the AI Provider - a nice property when you're still testing.

Actually, one honest caveat: this node's classification is deterministic over tags and the optional DB, so don't expect it to "understand" your prompt the way an LLM would. Its power is normalization and structure, not semantics. The moment you want real understanding, that's what the provider-driven nodes are for.

Install and gotchas

It's the same one-time install for the whole pack:

git clone https://github.com/GALIAIS/GALIAIS-Nodes.git ComfyUI/custom_nodes/GALIAIS-Nodes

Restart ComfyUI (or install "GALIAIS-Nodes" via ComfyUI Manager). Zero pip dependencies. One thing to know: if you want the analyzer to normalize tags against the actual Danbooru dictionary, you need the separately-maintained Danbooru-Tag-Database repo and the pack's DB loader - the database is not bundled because it's huge. The pack is new and quiet online, so expect to lean on the README's workflow diagram and the 分类摘要 output to figure out what's happening on your first run.