Hey devs
I’ve been working on Obsidian Neural, a VST3 plugin that integrates Stable Audio Open into music DAWs for real-time generation.
Technical approach:
- LLM generates contextual prompts from user input
- Stable Audio Open generates the audio
- VST handles MIDI triggering, tempo sync, and sample management
- Cloud inference API + self-hosted options
The goal: make AI audio generation playable rather than just generative.
Challenges I’m tackling:
- Real-time latency optimization (~10s generation)
- Context-aware prompt engineering for better results
- Integration with existing music workflows
You can discover it on GitHub at innermost47/ai-dj
Open source (AGPL v3.0) — all code is public.