flux prompt word generator

Getting an LLM to write good Flux prompts is mostly a prompt-engineering problem, not a Flux problem - you have to tell the LLM what kind of image it's writing for, in language specific enough that the output doesn't come out generic. flux_persona is a shortcut for that: pick a genre, fill in a theme and a style, and it hands you back a ready-made system prompt tuned for that genre, instead of you writing one from scratch every time.

It doesn't call an LLM and it doesn't touch Flux directly - it's a text template. What comes out is a system_prompt_input string you wire into this pack's LLM node, which then does the actual prompt-writing.

How it works

flux_persona is one of several "persona" nodes in this pack (this batch also covers ic_lora_persona) - think of the category as a library of pre-written system-prompt templates for specific creative jobs, so you're not starting from a blank text box every time you want the LLM to write in a particular register.

The inputs and outputs that matter

• image_type - which genre template to use. The choices sampled from the schema: Poster, Hearthstone (trading-card style), Yu-Gi-Oh (a different trading-card style), tarot, and manga. Each one steers the resulting system prompt toward that genre's visual conventions - card-frame composition for Hearthstone/Yu-Gi-Oh, mystical iconography for tarot, panel/line-art language for manga.
• theme (default: "party dog") - the subject of the image. Swap this for whatever you're actually generating.
• style (default: "fantasy") - a style descriptor layered on top of the genre.
• strength (optional, 0–10, default 5) - how strongly the theme/style get emphasized in the generated prompt. Push it up if the LLM's output keeps drifting away from what you asked for; the default is a middle setting, not a ceiling.
• is_enable - standard bypass toggle.

Output: system_prompt_input (STRING) - a template string. It's not a finished image prompt by itself; it's instructions for the LLM that will write the image prompt.

How to install it

Via ComfyUI Manager: search comfyui_LLM_party, install, restart. Manually:

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

Then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, and restart. This node needs nothing beyond the base pack install - no models, no API keys of its own - since it's just returning a string. The heavier requirements (local LLM support, GGUF loading) belong to other nodes in the pack; if you only want this one and its downstream LLM call, you can leave fast_installed at False and skip nothing, or set it True in config.ini if you're not touching GGUF models elsewhere.

Common issues & troubleshooting

Nothing happens, or the graph errors downstream. This node produces text, not an image - it has to feed into an LLM node (which writes the actual Flux prompt), which then feeds into your Flux sampling chain. If you wire system_prompt_input directly into something expecting a finished image prompt, you've skipped the LLM step this node exists to set up.

The output prompt doesn't match your theme. Double-check theme and style are actually filled in with what you want - the node's own defaults ("party dog", "fantasy") will happily produce exactly that if you forget to change them.

Results feel generic despite a specific image_type. This node only sets up the instructions; how well the downstream LLM follows them still depends on which model you've wired up. A weaker or more literal-minded model will lean less into the genre conventions than a strong instruction-following one.