V4 Base Prompt ✒️🅝🅐🅘

NovelAI's V4 and V4.5 models prompt differently than V3-era anime models: instead of a flat tag soup, the API expects a structured prompt where the main description lives in a base_caption field. V4BasePrompt is the pack's way of giving that field a dedicated box. It's deliberately boring - one multiline text input called base_caption, one STRING output - and its entire job is to keep your V4 positive prompt organized and obviously-placed before it feeds GenerateNAID.

The V4/V4.5 workflow the README documents is exactly this shape:

V4BasePrompt ------> positive
                          GenerateNAID
V4NegativePrompt -> negative

So the node isn't doing anything clever at generation time - its convert function literally returns the string you typed. What it buys you is clarity: when your GenerateNAID node is buried in a big graph, a dedicated "V4 base caption" box makes the V4-style prompting explicit instead of hiding it inside a generic positive-prompt field. NovelAI V4 is built around natural-language-ish captions plus tags, and the KB's prompt-engineering essay makes the same point about modern anime models: the base caption is where your subject, setting, and quality tags go, and it's worth treating as a distinct block.

There's one mechanism subtlety worth knowing, and it explains why this node exists at all: GenerateNAID wraps whatever lands in its positive input into the API's v4_prompt.base_caption field on every request anyway. So functionally, V4BasePrompt is a labeled passthrough whose output could be wired straight into positive - which is precisely how the README shows it being used. If you're wondering "why not just type in the Generate node," you can. The V4 nodes are for workflows that want the structure front and center, or that will later grow character captions.

And a heads-up so you don't get burned: this STRING output is not a NAID_OPTION chain link. It feeds the prompt inputs on GenerateNAID (the README's own wiring), not the option socket - connect it to option and nothing will happen. It also works fine with the older models; those just ignore the V4 framing.

Setup is the pack standard: ComfyUI Manager → search "ComfyUI_NAIDGenerator", or clone into custom_nodes/ and restart. It's a pure local text node - no Anlas, no API call, and it works even before you've configured your NovelAI token.