Master Key (Creepybits)

Master Key is the creative starting point of the Creepybits suite - the node the author calls the "Idea Engine." It's a text box where you type your raw idea - "boy, dog, park, red ball" or "a rainy cyberpunk alley" - and out comes that idea wrapped in a massive, hand-written system prompt that turns any downstream LLM into an expert prompt engineer.

That's the entire mechanic, and it's worth being blunt about it: the node itself is a string concatenation. It reads a long instruction file (prompt.txt in assets/prompts) and prepends it to whatever you type into text_2. The magic is in that file, which is a carefully-worded brief optimized for writing image-generation prompts for Flux-class models (it even name-checks "t5-xxl" and the Gemini 2.0 model line). You're not supposed to see the instructions - they're the seasoning, not the meal.

How it works

At startup the node loads prompt.txt once and caches it. Every run it joins that fixed text with your text_2 input and outputs the combined string as text. No API calls, no tokens spent, nothing executed - which is exactly why it's the safe first node in a graph: it can't fail, and it always produces output.

The two supported patterns from the docs:

• Keyword expansion - drop in a handful of keywords, get a fully fleshed scene brief from the LLM.
• Concept expansion - drop in a full sentence, get cinematic detail, emotional depth, and style direction.

Either way the wiring is the same: text output → system_prompt input of the pack's Gemini API node → generated prompt → your CLIP encoder.

The inputs that matter

Just one, technically: text_2, a multiline text box. There are no wired inputs. The node is a start node - you type or paste into it. The single text output is what you connect downstream.

Where it goes wrong

Because the bundled instructions are Flux-oriented ("optimized for... Flux AI with t5-xxl clips"), the generated prompts are tuned for that model family. On SDXL or other models they'll still work - good descriptive prompts generally transfer - but you may notice the vocabulary skewing Flux-flavored. If you're on a different model line, open prompt.txt and edit it. That's a feature, not a hack: the file is meant to be rewritten for your stack.

And like the other persona nodes, this one doesn't do the "master key" part itself. No LLM downstream, no expanded prompt - just your idea glued to instructions.

Installing it

Part of ComfyUI-Creepy_nodes:

• ComfyUI Manager: search "Creepy_nodes" and install.
• Manual:

  cd ComfyUI/custom_nodes
  git clone https://github.com/Creepybits/ComfyUI-Creepy_nodes.git
  

Restart ComfyUI. No dependencies for this node on its own - its partner Gemini node needs google-genai from the pack's requirements.

Troubleshooting

If your output starts with ERROR: Prompt file not found., the prompt.txt file has been moved or renamed in assets/prompts - restore it from the repo. If your generated prompts feel generic, the file loaded fine but the instructions don't match your model - edit prompt.txt. And if text_2 is empty, you still get output (just the instructions), so an empty box doesn't error - it just produces a less useful result.