stable audio 3.stub

---

id: stable-audio-3 name: Stable Audio 3 repo_path: ~/Documents/sw30labs/repos/stable-audio-3 remote_url: https://github.com/sw30labs/stable-audio-3.git primary_language: Python framework: PyTorch / Gradio category: local-inference-mlx stacks: [cli] ingested: 2026-05-29 last_commit_date: 2026-05-27

Purpose

Open-source platform for fast, high-quality AI-generated audio and music. Supports four model tiers: Small-Music (433M, CPU), Small-SFX (433M, CPU), Medium (1.4B, GPU/CUDA, 380s), and Large (2.7B, API-only). Built on lessons from stable-audio-tools; streamlined for inference and fine-tuning.

Architecture

• Models: SAME autoencoder family; small models CPU-only, medium requires CUDA GPU
• Gradio UI: run_gradio.py for interactive audio generation
• Optimized variants: optimized/ directory for quantized/efficient model paths
• Technical report: arxiv.org/abs/2605.17991

Dependencies

• PyTorch (CUDA for medium/large)
• Gradio
• Stability AI model checkpoints (HuggingFace)

CI/CD

Test suite present (tests/).

Entities and topics

• local-inference-mlx
• Tags: [local-inference, audio, llm]