🎵 Real-time AI music generation VST3 plugin for live performance
OBSIDIAN-Neural transforms AI into a live music instrument using intelligent LLM prompts and Stable Audio Open to create contextually aware loops directly in your DAW.
🚨 FRUSTRATED WITH INSTALLATION? 🚨
Skip the complexity! Get a free API key instead.
No Python, no compilation, no debugging - just music creation.
OBSIDIAN-Neural is currently in active development. This is NOT an official v1.0 release. While functional, you may encounter bugs and instabilities. I'm working on this continuously and push updates regularly (sometimes multiple times per day).
I may have been too enthusiastic about the installation complexity initially. I'm aware it can be frustrating, and I'm actively working to improve the user experience.
Get FREE API access - No setup required!
- Contact me for a free API key (only 10 slots available, unfortunately I can't provide more from my side)
- If you have a GPU server and want to share: the server works with API keys - you can host your own instance!
- Download VST3 from Releases
- That's it! No GPU, no Python, no compilation needed
- Download the .exe installer
- One-click setup (still requires GPU setup)
- Can be combined with manual VST3 copy (uncheck "Build VST3" in installer)
- Full source compilation (complex!)
- Only recommended if you're comfortable with Python, CMake, and GPU toolchains
TL;DR: AI models are MASSIVE and need serious hardware
- Stable Audio Open: ~4.8GB model size
- VRAM needed: 6+ GB minimum
- Loading time: 30-60 seconds
- A normal VST: 50MB, loads instantly
Every AI music tool faces this:
- Suno: Server-based
- Udio: Server-based
- Google MusicLM: Server-based
- Adobe AI Audio: Server-based
✅ Lighter VST - Fast loading, stable performance
✅ Shared resources - Multiple users, one GPU
✅ Better hardware - Dedicated inference machines
✅ Future-proof - Easy model updates
I'm aware that requiring expensive GPUs creates barriers. Many musicians can't afford RTX 4090s just to try AI music generation. The subscription model everywhere is getting ridiculous.
I'm researching a distributed GPU network where we can:
- Pool our computing resources
- Share GPU power across the community
- Provide completely free access to all users
- Redistribute donations to GPU contributors
- Build something truly community-owned
Got a powerful GPU and want to help? The server works with API keys - you can host your own instance and share access with the community!
Know a better way to handle massive AI models? Have ideas? Please share!
I'm open to alternatives, but client/server is currently the most practical solution for real-time AI music generation.
Testing Period: Daily 10:00 - 00:00 (French Time)
Only 10 free API keys available for motivated testers - no GPU required! Generate up to 50 AI samples using my hosted Stable Audio Open Small model.
- Bug reports via GitHub Issues
- Edge case testing and unusual workflows
- User experience feedback
- Clear documentation of any problems
Contact me at: b03caa1n5@mozmail.com
Include:
- Brief intro about your music/testing background
- What you'd like to explore with OBSIDIAN-Neural
- Commitment to provide constructive feedback
- Check Releases regularly
- Always download the highest version number
- I fix bugs as soon as I find them and push updates frequently
- Some releases may be unstable, but newer = more fixes
If using the installer + manual VST3:
- Download both installer and VST3 from Releases
- Run installer, uncheck "Build VST3"
- Manually copy VST3 to
C:\Program Files\Common Files\VST3\
📰 Press coverage moved to PRESS.md
- LLM Brain: Analyzes sessions and generates smart prompts
- Stable Audio Open: High-quality electronic music generation
- Real-time: Everything happens live for performance use
- Auto tempo sync: Matches your DAW automatically
- 8 independent tracks with MIDI triggering (C3-B3)
- Advanced waveform editor with zoom and precise loop points
- Smart time-stretching with DAW tempo sync
- Individual outputs for separate mixing
- 16-step programmable sequencer per track
- Perfect DAW sync - locked to host tempo and transport
- Real-time pattern editing while playing
- Seamless live performance with measure-boundary start/stop
- MIDI Learn system for hardware controllers
- Session save/load functionality
- Background processing - works with VST window closed
- Global custom prompts shared across projects
Click to watch: World's First AI Jam Partner in action!
If you choose the full installation:
- NVIDIA GPU with CUDA support (RTX 3060+ recommended)
- 8GB+ VRAM minimum
- 16GB+ system RAM recommended
- Python 3.10 specifically (other versions may fail)
- NVIDIA CUDA Toolkit
- CMake (3.16+)
- Git
- Visual Studio Build Tools (Windows) or GCC/G++ (Linux)
- Request access on Hugging Face
- Install HF CLI:
pip install huggingface_hub - Login:
huggingface-cli login - Enter your token when prompted
License: Free for personal use and commercial use under $1M revenue/year
- Setup: Start AI server, load OBSIDIAN-Neural in DAW
- Generate: Add tracks, enter creative prompts, generate AI audio
- Perform: Trigger tracks with MIDI, auto-sync to DAW tempo
- Control: Map hardware controllers with MIDI Learn
Example LLM-generated prompt:
"Deep techno kick with sidechain compression, 126 BPM, dark atmosphere, minimal hi-hats, rolling bassline"
Found issues? You're helping improve OBSIDIAN-Neural for everyone!
- Create GitHub Issue
- Include:
- DAW name/version
- Operating system
- Steps to reproduce
- Expected vs actual behavior
- Screenshots if UI-related
- Crashes or data loss
- Timing/sync problems
- AI generation failures
- MIDI mapping issues
✅ VST3 Plugin: Fully functional for live performance
🔄 Installation: Working on simplification
🌍 Community: Building distributed GPU network
Star count since June 6th: 53 ⭐ - Thank you for the support!
MIT License - Feel free to modify, but please keep attribution to InnerMost47
Music & Creative Coding:
- YouTube Channel - Original compositions
- Unexpected Records - Mobile recording studio
- Randomizer - Generative music studio
AI Art Projects:
- AutoGenius Daily - AI personas platform
- AI Harmony Radio - 24/7 experimental radio
OBSIDIAN-Neural - Where artificial intelligence meets live music performance.
Developed by InnerMost47