Translate Persona

LLM Party's "persona" nodes exist so you don't have to hand-write a system prompt every time you want a model to act as a specific role. They all output a system_prompt_input string, which is the socket name this pack's LLM loader nodes use for the system prompt - so any persona node here is meant to plug straight into that. translate_persona builds the specific persona of a two-way translator: give it a language pair and a couple of style knobs, and it assembles a system prompt configuring the model to translate between them, instead of you writing "You are a professional translator..." by hand every time.

Inputs and outputs that matter

• language_A / language_B - the two languages to translate between, defaulting to Chinese (中文) and English (英文) since that's the pack's origin. Swap in whatever pair you actually need.
• tone (optional, default 正式/"formal") - the register the translation should aim for. Change it if you want something casual, or match the tone of the source material.
• degree (optional, INT, max 10, default 5) - a strength/intensity dial. The schema doesn't spell out exactly what it scales, but on a translator persona, this reads as controlling how liberal versus literal the translation leans - worth experimenting with at the extremes to see how your particular model responds to it.
• is_enable - the pack-wide skip switch.

The single output, system_prompt_input, is the assembled prompt string - wire it into your LLM node's system prompt socket, not into the user message.

Installing it

Search comfyui_LLM_party in ComfyUI Manager and install, or clone directly:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party

pip install -r requirements.txt inside your ComfyUI Python environment, then restart. This node itself has no model or API dependency - it just builds text - but it's bundled inside the full pack, which pulls in a much larger set of dependencies overall (LLM API clients, local model loading, RAG, TTS, and more). The README's only_api branch is the lighter option if you only need API-based calling.

Common issues

This node configures the system prompt; it doesn't configure which model is doing the translating, or how good that model actually is at the language pair you picked. If translation quality is poor, that's a model choice and prompt-tuning problem downstream, not something translate_persona itself can fix - some models are meaningfully better at certain language pairs than others, and no system prompt closes that gap entirely.

Also worth remembering: system_prompt_input replaces the role the model is playing for that conversation. If you're combining this with other persona nodes or a custom system prompt elsewhere in the same LLM chain, only one system prompt generally wins - check how your specific LLM node handles multiple incoming system-prompt-shaped inputs before assuming they'll layer together cleanly. If you want the model to be a translator and hold a persona at the same time, that's usually better handled by writing one combined prompt rather than stacking two persona nodes and hoping they merge.