StableAudioKit 0.0.0-rc1

StableAudioKit v0.0.0-rc1

Assets

• StableAudioKit-0.0.0-rc1.xcframework.zip — multi-platform XCFramework (macOS, iOS, iOS Simulator, visionOS, visionOS Simulator). Statically bundles mlx-swift and swift-sentencepiece; each slice ships a precompiled mlx.metallib.
• StableAudioCLI-0.0.0-rc1-macos-arm64.tar.gz — StableAudioCLI for macOS arm64 with the colocated MLX metal library.
• prepare_weights_standalone.py — single-file Python helper for downloading and converting the gated Stability AI weights (PEP 723 inline-script — run with uv).

Using the CLI

1. Download and unpack:

   tar -xzf StableAudioCLI-0.0.0-rc1-macos-arm64.tar.gz
   cd StableAudioCLI-0.0.0-rc1-macos-arm64

2. Unblock Gatekeeper. This binary is unsigned and un-notarized. macOS will quarantine it on download. Clear the attribute before launching:

   xattr -dr com.apple.quarantine StableAudioCLI mlx.metallib

3. Keep mlx.metallib next to StableAudioCLI — the binary looks for the metal library beside itself at runtime.

Preparing weights

The Stable Audio 3 weights are gated. Request access on Hugging Face first:

• https://huggingface.co/stabilityai/stable-audio-3-optimized

Then authenticate either with hf auth login or by exporting HF_TOKEN=....

The recommended way to run the prep script is via uv — it reads the PEP 723 header inside prepare_weights_standalone.py and provisions an isolated environment automatically:

# One-time: install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Authorize (either form)
hf auth login                  # interactive
export HF_TOKEN=hf_xxx          # non-interactive

# Download and convert
uv run prepare_weights_standalone.py \
  --download \
  --variants small medium \
  --with-encoder \
  --destination ./Weights

Then point the CLI at the output directory:

./StableAudioCLI \
  --model smallMusic \
  --prompt "lofi house loop, 120 BPM" \
  --duration 10 \
  --model-path ./Weights \
  -o output.wav

Using the XCFramework

Drop StableAudioKit.xcframework into any Xcode project targeting macOS 14+, iOS 17+, or visionOS 1+. See the repository README for the public StableAudioPipeline API.