Nodes/world weaver/Master Key (World Weaver)
ComfyUI Node

Master Key (World Weaver)

The node that bolts a system prompt onto your idea

By Creepybits·Created 10 months ago·Updated 2 months ago· 30
Master Key (World Weaver)
    • text
    text_2

    Master Key is the pack's front door for text. You type a character description into it, and it returns that text with a big, fixed system prompt glued onto the front. What comes out is a fully-framed instruction for whatever LLM or CLIP encoder you point it at next. It's the node the pack's Genesis tester - the "should I even keep this character?" workflow - runs your draft prompt through before you commit it to the database.

    The name is drama; the mechanism is a file read plus a string concat. Master Key loads a text file from the pack (assets/prompts/prompt.txt) once at startup, then simply prepends it to whatever you type in text_2. One input, one output:

    • text_2 - your character description, multiline.
    • Output: text - the system prompt followed immediately by your text.

    What's actually in that system prompt

    The file it reads is worth knowing about because it's opinionated in ways that will leak into your results. It's tuned for Flux with the T5-XXL text encoder and Gemini-era prompting: it demands a 200–400 token output, bans describing the result as a painting or digital art (photoreal or nothing), insists on camera and lighting language, and - the notable part - includes an explicit "no censorship" clause aimed at Gemini. It also carries conditional video logic: if your text implies motion or a "cinematic shot," the prompt switches from photography mode to cinematography, banning static terms and demanding camera movement and environmental physics. That's why the pack's Wan 2.2 video workflows run their descriptions through this node.

    Practically, that means Master Key is not a neutral text prepender. It's a specific recipe for a specific model family. If you're generating with Flux, that recipe is probably in your favor. If you're on a booru-tag SDXL model or a 2026 instruction-encoded base, an LLM-style prose system prompt can be inert or worse - the prompt engineering rules changed with the text encoder, and a photo prompt doesn't carry across the whole family tree.

    Installing it

    Standard pack install: ComfyUI Manager, search "world weaver", install, restart. Or:

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

    Restart, let the dependencies install (google-genai, numpy, opencv-python, requests, sounddevice - no model downloads).

    Where people get burned

    The classic one: the fixed text is loaded when the node is created, so if you edit prompt.txt to tune the system prompt, you need to restart ComfyUI (or at least recreate the node) before your edit takes effect. Editing the file and wondering why nothing changed is the standard facepalm.

    Also note that the output is a single string with no separator marker between system prompt and your text - if you're sending this to a chat-format LLM later, that boundary matters and this node won't draw it for you. It's built for the pack's own plumbing, where the Gemini node gets its system prompt separately and Master Key's output is really the user content in disguise. If that sounds redundant, it is, a little - but it's also the simplest way World Weaver has to make sure every generation gets the same, careful, Flux-optimized framing.

    CategoryCreepybits/World_weaver

    Inputs (1)

    NameTypeDefaultDescription
    text_2STRING

    Outputs (1)

    NameTypeDescription
    textSTRING