OBSIDIAN Neural just integrated Stable Audio 3 Medium, Stability AI's latest open-weight audio model, as the 9th engine in the plugin.
SA3 brings a new architecture to the lineup: FlowMatching with 8-step inference, making generation significantly faster than the diffusion-based models already supported. Early tests have been particularly convincing on world instruments (sitar, oud, koto, shakuhachi, panflute...).
Like the other 8 engines, SA3 runs on the distributed GPU provider network, with cloud fallback via fal.ai when no node is available. Available right now in the latest version, free and open source.
What's new with this release:
- Stable Audio 3 Medium added as the 9th selectable AI engine.
- Optional RAM caching on provider nodes for faster cold starts on SA3.
- Cross-platform as always: Windows, macOS (Intel + Apple Silicon), Linux.
Links:
- Plugin: https://github.com/innermost47/ai-dj
- Provider kit: https://github.com/innermost47/obsidian-neural-provider
- Website: https://obsidian-neural.com
OBSIDIAN Neural was recently presented at the AES 160th Convention 2026 in Copenhagen (Engineering Brief) and at the AES AIMLA 2025 in London (Late Breaking Demo Paper, Queen Mary University).
