Load Persona

comfyui_LLM_party ships with a library of premade system prompts, and this is the node that hands you one from a dropdown instead of you writing your own from scratch. The pack calls this concept "persona" - the README literally frames the whole project around it, describing everything "from role setting to quickly build your own exclusive AI assistant" as a core use case, and Load Persona is where that starts.

The mechanism is simple: pick a name, get a ready-made system prompt string out. No API call, no external service - the personas are bundled with the pack itself. What's actually interesting is what's in the list. The schema shows 21 choices total, and the sample includes things like canvas, a "FLUX prompt generator," HTML2IMG, an "LLM tool maker," GPT_DAN_ZH, two knowledge-graph-flavored personas (csv and json variants), a "Chinese new interpretation" persona, and one called omost. That last one isn't a coincidence - the README's own credits section explicitly lists lllyasviel's Omost project as one of two projects this pack borrowed nodes from, and Omost's whole idea is using an LLM to plan out detailed, region-by-region image composition before generation. Seeing it show up as a selectable persona here is the same lineage showing through: pick it, and you're loading a system prompt built to make an LLM think like Omost's planner, feeding structured prompts into an image workflow rather than freeform chat.

That FLUX-prompt-generator persona and the image-prompt-generator entries point at the same pattern from a different angle - this pack isn't only for chatbots, it's explicitly built to bridge LLM output into image generation, and personas are one of the mechanisms for that. Set the right persona, and the LLM node downstream isn't answering questions, it's writing prompts for whatever comes next in your image pipeline.

Inputs that matter. persona_name is the dropdown - required, no default value shown, so you do need to actually pick one rather than relying on a fallback. is_enable is this pack's standard toggle to switch persona injection off without removing the node from the graph. There's also an optional file_content field with no documented tooltip - worth treating as a way to layer your own text on top of a chosen persona rather than something required for basic use; if you just want one of the 21 presets as-is, leave it blank. The single output is system_prompt_input, a STRING, which wires directly into your LLM node's system prompt input.

Installing it. Load Persona ships with the full comfyui_LLM_party pack - no separate install. Get it through ComfyUI Manager (search "comfyui_LLM_party") or cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then pip install -r requirements.txt from inside the project folder using your ComfyUI Python environment, and restart. The personas themselves are bundled files, not downloads, so once the pack installs correctly, all 21 are available immediately with nothing further to fetch.

Troubleshooting. If the dropdown looks empty or shorter than expected, that's almost always an incomplete install rather than anything wrong with this node specifically - the persona files ship inside the pack's own folder structure, and a partial clone or a failed git pull can leave some of them missing. If the whole node set fails to load rather than just personas looking thin, that's this project's much larger requirements.txt - real users have reported the entire pack failing to import after a fresh install, and it's a dependency conflict, not this node - the README keeps requirements_fixed.txt around as the fallback worth trying first.