NAI 标签建议

NovelAI's own site has a tag-suggestion feature that finishes your prompt with relevant tags while you type. This node is that feature, exposed inside ComfyUI. You give it a seed prompt and it asks NovelAI what you probably meant, which is handy when you know the general idea but not the exact Danbooru vocabulary for it.

What it is

Inputs are an NAI_PIPE (it needs the model selection to know which tag set to suggest against), a prompt string (default 1girl), and a limit (1 to 50, default 10) controlling how many suggestions come back. It calls NovelAI's suggest-tags endpoint and outputs a single 标签建议JSON (STRING) - a JSON array of suggested tags.

So the flow is: type 1girl, school, run the node, and read what NovelAI thinks belongs in that prompt. Then paste the suggestions you like into your prompt node and run for real. It's a research tool, not a generation node - it spends nothing itself.

How it behaves

The suggestions come from NovelAI's own model vocabulary, which is the whole value. The KB's prompt-engineering notes hammer on this: Danbooru-tag models respond to their actual tag vocabulary far better than to natural-language descriptions, and the gap between "what I described" and "what the model knows" is where most frustration lives. This node bridges that gap by asking the model directly.

Two honest caveats:

• Suggestions are suggestions. They're the model's prior on what completes a prompt, which means they lean generic - you'll get blue eyes, long hair, smile, not the weirdly specific niche tag that makes a prompt sing. Use them to fill gaps, not as a creative oracle.
• The output is JSON, not a formatted prompt. It's meant to be read, not wired straight into a sampler. You'll be copying strings around by hand.

Does it cost anything?

The suggest-tags call is free - no Anlas for suggestions themselves. That's worth knowing because the surrounding pack is otherwise pay-per-call, and this is one of the few nodes where you can experiment without opening your wallet.

Install and setup

Ships in ComfyUI-NovelAI-GENYTOOLS. Manager search, or:

cd ComfyUI/custom_nodes
git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git
pip install -r ComfyUI-NovelAI-GENYTOOLS/requirements.txt

Restart, set your NovelAI Persistent API Token under NovelAI 2.0: Persistent API Token, and you're done. No models to download.

Gotchas

• Needs the NAI_PIPE for a reason: pick a different model and you may get different tag suggestions, since the vocabulary tracks the model. Don't wire in a model you don't intend to generate with.
• JSON output means no pretty display by default - throw it into any text/JSON viewer node.
• Free as of writing, but it's still an API call to NovelAI. If NovelAI ever meters it (they've been moving things toward Anlas over time), this node's cost story changes without the node itself changing.