Skip to content

stability.md

Latest commit

 

History

History
118 lines (86 loc) · 3 KB

stability.md

File metadata and controls

118 lines (86 loc) · 3 KB
Raw

Stability AI Provider

ORAM can use Stability AI Stable Audio for text-to-music, text-to-sound, audio-to-audio, inpaint, continuation, and LoRA-conditioned material generation through the local engine registry.

Configure In The macOS App

  1. Open Provider Setup.
  2. Paste the Stability AI API key.
  3. Save to Keychain.
  4. Pick Stable Audio in the generator selector.

The app writes the key directly to:

service: com.sonicfieldlabs.oram
account: provider:stability

Configure In The Terminal

oram credentials set stability
oram credentials status
oram credentials test stability

Then select the engine explicitly when using daemon/app generation:

stability-stable-audio-3

The older Stable Audio 2.5 adapter remains available as:

stability-stable-audio-25

Stable Audio 3 Modes

ORAM exposes these SA3-oriented daemon modes:

ORAM Generate     -> prompt to clip
ORAM Morph        -> source layer plus prompt with noise_depth
ORAM Continue     -> extend source layer; selected loop region can guide range
ORAM Inpaint      -> replace selected range or layer loop region
ORAM LoRA Mixer   -> prompt generation with LoRA A/B stack metadata
ORAM Latent       -> local sidecar mode for SAME latent workflows

Apps and external clients can inspect modes with:

GET /stable-audio/modes

Render through ORAM-owned daemon layers:

POST /stable-audio/render

Render for DAW, Max, or standalone clients that own their own audio state:

POST /plugin/stable-audio/render

Developer Fallback

.env remains supported for development:

# STABILITY_API_KEY=<developer-key>
# Optional: override the Stability API route if the public SA3 endpoint changes.
# ORAM_STABLE_AUDIO_API_URL=https://api.stability.ai/...

Keychain is preferred for packaged app usage.

Local Stable Audio 3 Service

For Apple Silicon MLX or a local Python/CUDA sidecar, point ORAM at the local service. ORAM defaults to the GERM-compatible local server URL when loaded from the environment:

ORAM_STABLE_AUDIO_SERVICE_URL=http://127.0.0.1:8765
ORAM_STABLE_AUDIO_LOCAL_PROVIDER=stable_audio_mlx
ORAM_STABLE_AUDIO_LOCAL_MODEL=medium-mlx
ORAM_STABLE_AUDIO_DECODER=same-s

ORAM sends a JSON payload to /render first. If that route is not available, it uses GERM mode routes:

generate     -> /generate
morph        -> /audio-to-audio
inpaint      -> /inpaint
continue     -> /continue
lora_mixer   -> /generate with LoRA metadata

Responses may be raw audio, base64 audio, a local audio_path, a JSON object containing an audio URL, or a GERM audio_files result served through /files/.... Model rendering must stay in this companion service, not in plugin or realtime audio callbacks.

Privacy

Stability AI receives the prompt and generation parameters needed for Stable Audio when the Stability API engine is selected. Local service requests stay on the configured service URL. ORAM does not send the key to Sonic Field Labs or any ORAM-operated server.