AEmotionStudio/foundation1-models

Foundation-1 — Mirror

BPM/Key-Aware Music Sample Generator

Original Model by RoyalCities · Fine-tuned on stable-audio-open-1.0

This is an ungated mirror of the Foundation-1 model weights for use with ComfyUI-FFMPEGA. All credits go to the original authors.

What's in This Repo

File	Description	Size
Foundation_1.safetensors	FP16 model checkpoint	~3 GB
model_config.json	Model architecture config	~1 KB

What Foundation-1 Does

Foundation-1 generates production-ready musical loops with fine-grained control over:

• Musical structure: BPM, bars, time signatures, key/mode
• Instrument identity: 30+ instrument families (synth, strings, brass, guitar, etc.)
• Timbral control: 100+ timbre tags (warm, bright, gritty, glassy, etc.)
• FX prompting: Reverb, delay, chorus, distortion, and more
• Loop fidelity: Seamless, tempo-synced loops designed for layering

Usage with ComfyUI-FFMPEGA

These weights are auto-downloaded by the FFMPEGA Agent node:

1. Set llm_model to none
2. Set no_llm_mode to generate_sample
3. Enter a prompt like: "Synth, Pad, Warm, Wide, Lush, 120 BPM, 4 Bars, C major"

The model (~3 GB) downloads on first use to ComfyUI/models/foundation1/.

Hardware Requirements

• VRAM: ~7 GB during generation
• Generation speed: ~7–8 seconds per sample (RTX 3090)

Prompt Structure

Foundation-1 uses a layered prompt system:

[Instrument Family], [Sub-Family], [Timbre Tags], [FX Tags], [BPM], [Bars], [Key]

Example: "Synth, Lead, Bright, Sharp, Saw, Detune, Delay Ping Pong, 140 BPM, 8 Bars, A minor"

License

These model weights are released under the Stability AI Community License.

• ✅ Free for non-commercial use
• ✅ Free for commercial use by entities with annual revenue < $1M USD
• ❌ Entities with annual revenue ≥ $1M need an Enterprise license from Stability AI

Acknowledgements

• Foundation-1 by RoyalCities
• Stable Audio Open by Stability AI