SCG Foundation-1 Random Prompt

The hardest part of picking up a new text-to-sample model isn't learning the nodes - it's having any idea what the thing actually sounds like. The SCG Foundation-1 Random Prompt node is the slot machine for exactly that. It generates a structured prompt from Foundation-1's own tag vocabulary, using the weighted random system from the RC Enhanced Fork that the model is based on, and hands you a STRING to feed the Sample Generator.

It's the node you reach for when you don't have a target in mind and just want to audition the model's range, or when you want a batch of varied loops to queue up. The Prompt Builder is for when you know what you want; this is for when you don't.

Simple vs Experimental

• Simple (default) - one instrument family, a weighted sub-family pick, 2–4 timbre tags, maybe a spatial or wave tag. Coherent and tasteful; it sounds like one instrument playing a loop. Best when you actually intend to use the result.
• Experimental - the richer mode. More timbres (4–7), more spatial/wave/style tags, and for the Synth and Bass families it throws in character like Acid/303 styles and FM textures. Less predictable, more interesting, occasionally weird. Good for discovering combinations you'd never type yourself.

The weights are the point of "weighted": within a family, sub-families are drawn with real probability weights (Synth Lead at 40 vs Atmosphere at 4), so you get plausible picks instead of uniform dice rolls. Same for the FX tags.

The inputs

• mode - Simple or Experimental, above.
• seed - -1 (default) means fresh randomness every run; any value from 0 to 2^31 makes the prompt reproducible. Note the random-prompt seed is separate from the Sample Generator's seed - that one controls the audio noise, not the tag pick.
• family_override - Random picks any family; or force one (Synth, Keys, Bass, …) to keep a genre consistent while everything else varies. This is the input you'll actually use: set it to Bass and mash the queue button to audition basses all afternoon.

One output, prompt, a STRING that wires straight into the Sample Generator's prompt input. The generator still appends key/scale/bars/BPM, so the random prompt stays pure tag text.

Installing, and the one trap

Standard pack install: ComfyUI Manager → search "SCG Foundation-1" → install → restart. By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/SanDiegoDude/scg_Foundation-1-comfyUI
cd scg_Foundation-1-comfyUI
python install.py

And the trap worth repeating because it will otherwise wreck your setup: stable-audio-tools must be installed with --no-deps, which install.py does for you. Running plain pip install stable-audio-tools or pip install -r requirements.txt first will force-downgrade torch, numpy, and pandas and break your ComfyUI environment. If something exploded mid-install, that's almost certainly why.

Where it sits

Random Prompt is prompt-generation only - it doesn't need the Loader to run, but the full chain (Loader → Sample Generator) has to be alive before its output does you any good. Practical workflow: feed Random Prompt into the sampler, set family_override to taste, and let the sampler's autoplay preview results as they land. It's the fastest feedback loop this pack offers.