RPG Character Selector

If you've ever spent twenty minutes typing "elf ranger with a distinctive hairstyle" and then re-rolling the whole prompt because the hair came out wrong, this is the node for you. RPGCharacterSelector is the heart of the ComfyUI-RPG-Characters pack: thirteen dropdowns that read like a character sheet (race, ethnicity, gender, age, class, hair, beard, clothes, augmentations, emotion, scene), and it turns your picks into a finished, structured prompt - positive and negative - aimed at 1024x1024 extreme close-up portraits.

It's a pure string generator. No model, no CLIP, no GPU work - you feed it dropdown choices and it hands you prompt text. That's the whole trick, and it's why it works with SD 1.5, SDXL, and Flux alike: it doesn't care what's downstream, it just writes better tags than you will at 2am.

How it works

Every dropdown maps to a Python dictionary of curated prompt fragments (the pack ships ~13 data files, each with dozens of entries). When you hit run, the node looks up each selection, joins the fragments, and prefixes the whole thing with "Create an extreme close-up portrait of". The negative prompt is assembled the same way from per-entry negatives.

Here's the sneaky part: a lot of entries contain {variant|variant|variant} blocks, and the node randomly picks one on every run. So "Elf" isn't one prompt - it's a family, and you get a different member each time. That's great for variety when you're batch-generating a tavern full of NPCs, and terrible if you're trying to reproduce a face. The selection_summary output is your fix: it lists every random pick that run, so you can see exactly what changed and copy it if you want it back.

Scene gets special handling. Plain White / Plain Black / Green Screen and the other "plain" scenes don't just become a location tag - they become "Scene: Solid X Background Only. No objects or distractions." That's a deliberate nod to chroma-key workflows, where a busy prompt will wreck your background removal.

The inputs and outputs that matter

You set the dropdowns. The only one worth flagging: Augmentations, which is the pack's sci-fi/cyberpunk bolt-on (Magic Tattoo, Cybernetic Eye, Subdermal Armor…). If you're doing pure fantasy, leave it on "No Augmentations".

Outputs:

• positive_prompt / negative_prompt - the standard tag strings. Wire these into a CLIP Text Encode (or the pack's own concatenator, which we'll get to).
• Ollama_Generate_V2_Textbox_2 - the same description, but phrased as prose for an LLM. It feeds ComfyUI-Ollama's OllamaGenerateV2 node if you run the local-LLM path.
• selection_summary - a plain-text record of your choices plus the random picks. Wire it to a text preview and you've got a free "what did I just generate" readout.

Install

In ComfyUI Manager, search RPG-Characters and install, or clone manually:

cd ComfyUI/custom_nodes
git clone https://github.com/lord-lethris/ComfyUI-RPG-Characters.git

Restart ComfyUI. There's no requirements.txt and no model downloads - this pack is pure Python. The only external dependency is optional: the Ollama outputs need a running Ollama server plus the stavsap/ComfyUI-Ollama node; skip that and the string outputs still work fine.

Common issues

• Outputs change between runs. That's the random variants, not a bug. Use selection_summary to track what happened.
• Nothing comes out of the Ollama outputs. The README is upfront: those need Ollama installed and running, and ComfyUI-Ollama installed, or they do nothing. The plain string outputs don't.
• The pack has a small footprint. This is a hobbyist pack from a single author, not a big ecosystem project - expect occasional rough edges and read the repo's Issues if something misbehaves.