blackbearreloaded/ProsperoAI-Stable-Audio-Open-Small-FP16-PS5

ProsperoAI Stable Audio Open Small FP16 for PlayStation 5

Powered by Stability AI.

This is the curated, install-ready Stable Audio Open Small FP16 folder for ProsperoAI, a native local AI application for PlayStation 5. It combines the Stability AI model weights with the pinned compatibility assets from a8nova/adreno-llms and was validated with ProsperoAI's native PS5 AGC GPU acceleration.

Install

Download this repository and copy the complete stable-audio-open-small/ folder to:

PPSA99004/models/stable-audio-open-small/

Keep the folder name and internal structure unchanged. Restart ProsperoAI, open Workshop, and select Stable Audio Open Small FP16. Playback starts when generation finishes. Saved audio can be played again from its conversation with Triangle.

Runtime profile

• Purpose: text to audio
• Architecture: T5 encoder, diffusion transformer, and Oobleck VAE decoder
• Precision: FP16
• Output: PCM16 stereo WAV at 44.1 kHz
• Current Alpha output: 16,384 frames, about 0.372 seconds
• Denoising: eight deterministic steps
• Acceleration: dense projections use the native PS5 AGC GPU path
• Prompt conditioning: CPU T5 encoder

The runtime presently favors deterministic validation over duration. A typical generation takes about 62 seconds on the tested PS5. This bundle does not yet expose the upstream model's full variable duration of up to 11 seconds.

PlayStation 5 validation

The curated folder generated finite stereo audio in the sandboxed PPSA99004 application on firmware 6.02. The validated WAV is 65,580 bytes with SHA-256 499d7bd51b98cb6bc0618b093ec06d7e0555d85bf14788124c790c07a42327a5. Its waveform matched the host oracle with cosine similarity 0.999908. The application archived, auto-played, restored, and replayed the WAV from a saved session.

This is an Alpha port. It is best suited to short sound experiments and is not yet a real-time or long-form music generator.

Reproduce this folder

provenance/model-preparation.json records every required file, pinned source revision, hash, generated decoder fold, runtime setting, and PS5 validation. With the ProsperoAI Alpha 2 model tools, run:

prepare-model \
  --provenance provenance/model-preparation.json \
  /path/to/pinned-stable-audio-source

The recipe verifies all 67 runtime files and publishes the folder only if the result matches the hardware-tested artifact.

License and attribution

The Stability AI model is distributed under the Stability AI Community License. See NOTICE for its required attribution and the packaging changes. The bundled adreno-llms compatibility files are covered by Apache License 2.0. Commercial users must follow the registration and revenue conditions in the Stability AI license and all use must comply with its Acceptable Use Policy.

ProsperoAI packaging and PS5 validation: BlackBearReloaded.