SCG Foundation-1 Sample Generator

ComfyUI nodes for Foundation-1 by RoyalCities — a structured text-to-sample model for music production. Generates tempo-synced, key-aware loops with deep control over instrumentation, timbre, FX, and musical structure.

Screenshots

Credits

This node wraps the Foundation-1 model created by RoyalCities.

• Model: RoyalCities/Foundation-1 on HuggingFace
• Built on Stability AI's stable-audio-tools
• Prompt system and tag vocabulary informed by the RC Enhanced Fork

All model weights are downloaded directly from the original HuggingFace repository. This project provides only the ComfyUI integration layer.

Installation

Do NOT run pip install -r requirements.txt by itself.

This node depends on stable-audio-tools, which pins dozens of training-only dependencies to old versions (torch 2.10, numpy 1.23, pandas 2.0, etc.) and will break your ComfyUI environment if installed normally.

Option 1 — ComfyUI Manager (recommended)

Install through the ComfyUI Manager UI. It will run install.py automatically, which handles the --no-deps install.

Option 2 — Run the install script

cd ComfyUI/custom_nodes/scg_Foundation-1-comfyUI
python install.py

Option 3 — Manual pip (two steps, in order)

pip install stable-audio-tools --no-deps
pip install -r requirements.txt

Nodes

Node	Purpose
SCG Foundation-1 Loader	Downloads the model from HuggingFace on first use (~1.5 GB) and loads it onto GPU. Caches in ComfyUI/models/audio_checkpoints/Foundation-1/.
SCG Foundation-1 Sample Generator	Main generation node. Set BPM, bars, key, scale, sampler params, and seed. Outputs ComfyUI AUDIO with built-in preview player. Supports negative prompts and audio-to-audio style transfer.
SCG Foundation-1 Prompt Builder	Build structured prompts from dropdowns — instrument family, sub-family, timbre, FX, spatial, structure, speed, density, and more.
SCG Foundation-1 Random Prompt	Weighted random prompt generation in Simple or Experimental mode, matching the RC Enhanced Fork's random prompt system.

Typical workflow

[Loader] ──→ [Sample Generator] ──→ [Save Audio / Preview Audio]
                     ↑
        [Prompt Builder] or [Random Prompt]

The prompt nodes output a STRING with instrument/timbre/FX/structure tags. The Sample Generator appends key, scale, bars, and BPM automatically and handles all timing alignment for loop-accurate output.

Supported parameters

Musical controls: 4 or 8 bars · 100–150 BPM · All 12 keys · Major/minor

Sampler: dpmpp-3m-sde (default), dpmpp-2m-sde, k-heun, k-lms, k-dpmpp-2s-ancestral, k-dpm-2, k-dpm-fast · 1–500 steps · CFG 0–25 · Sigma min/max

Style transfer: Connect any AUDIO input as init audio with adjustable noise level

Prompt tags: 10 instrument families · 40+ sub-families · 50 timbre descriptors · 11 spatial tags · 13 wave/tech tags · 11 style tags · Reverb/delay/distortion/modulation FX with graduated levels · Structure, speed, density, contour, rhythm controls

Hardware requirements

~7 GB VRAM during generation. A GPU with at least 8 GB VRAM is recommended.

License

The Foundation-1 model is licensed under the Stability AI Community License (non-commercial, or commercial under $1M revenue).