REDnote copywriting Persona📕

"Red Book" here is Xiaohongshu (小红书, "REDnote" in English), a huge Chinese social and e-commerce app - think Instagram crossed with Pinterest crossed with product reviews - where a very particular breezy, hashtag-heavy copywriting style genuinely outperforms generic marketing text. This node is a persona built specifically to write in that style. You're not describing "write me an ad"; you're loading a system prompt tuned for one platform's actual house voice.

It sits in this pack's persona category - nodes that shape how an LLM behaves rather than answer a one-off question - and it's a canned, fill-in-the-blanks template rather than something you write from scratch, the same general idea as this pack's Custom Persona node but pre-built for one very specific job.

How it works

You fill in a handful of fields describing what you want written, and the node assembles a ready-made system prompt and user prompt pair, tuned to Xiaohongshu's conventions, wired straight into this pack's LLM node.

The inputs and outputs that matter

• main_body (default "Red Book Text Persona") - the subject or product the copy is about.
• text_type - a two-way choice between the schema's own Chinese labels: 产品文案 ("product copy," for a specific item) and 活动文案 ("event/campaign copy," for a promotion or launch).
• min_length (default 100) - a floor on how long the generated copy has to be, so the model doesn't hand you three sentences when you needed a full post.
• desc - extra detail or direction for what should be covered.
• must_include_tag (optional) - a hashtag or keyword you want guaranteed to appear somewhere in the output, a product name or campaign tag.

Two outputs: system_prompt_input, the persona itself, and user_prompt_input, the specific request built from your fields - wire both into your LLM node's respective system and user prompt inputs.

How to install it

• ComfyUI Manager: search "comfyui_LLM_party", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, then restart.

Pure prompt templating - no external dependency of its own, nothing GPU-related.

Common issues & troubleshooting

Output comes back in Chinese even though you wrote desc in English. The persona's whole frame of reference is a Chinese platform, and its own field values (text_type's options) are Chinese - if you specifically want English copy, say so explicitly in desc rather than assuming the model will infer it.

must_include_tag isn't reliably showing up. This is a soft instruction to the model, not a hard constraint enforced after the fact - if it's being dropped, try making it a more prominent part of desc as well, or check the output and regenerate rather than assuming it's guaranteed on every call.

Copy comes back shorter than min_length. That's the model treating the floor as a rough target rather than a hard rule, which happens with weaker or faster models more than stronger ones - worth trying a better model_name upstream on the LLM node this persona feeds if length compliance matters.

Separate from this node's own behavior: comfyui_LLM_party has a real, recurring install issue where the whole pack fails to import after a Manager install, a dependency conflict rather than anything specific to persona nodes - check the ComfyUI console for an import error if this node doesn't show up at all.