🎵 Real-time AI music generation VST3 plugin for live performance
Base OBSIDIAN-Neural workflow tutorial (French with English subtitles available) - from DAW setup to live performance with hardware synths!
Video created by Chart Photography
💡 Enable English subtitles or English audio dubbing on YouTube for international viewers
🎯 Want to see YOUR jam sessions featured here?
I'm the only one posting live OBSIDIAN jams so far, and honestly... I'm dying to see how YOU use it!
Every AI-generated rhythm is different, sometimes weird, sometimes genius. That's the beauty - it forces you out of your comfort zone and into unexplored territory.
Share your OBSIDIAN videos:
📧 Email me: [email protected]
💬 Or create a Discussion with your YouTube link!
AI + analog gear jungle improvisation - when OBSIDIAN throws you curveballs and you roll with it!
📺 Check out all community jams → YOUTUBE.md
Let's build a community of AI-human musical chaos! 🤖🎵
📖 Read the full story - Why I built an AI that jams with you in real-time
🎉 Runs completely offline with local TensorFlow Lite models!
No more servers, Python, or GPU requirements!
- Get Stability AI access: Request access to Stable Audio Open on Hugging Face
- Download models: Get all files from innermost47/stable-audio-open-small-tflite
- Simple setup: Copy models to
%APPDATA%\OBSIDIAN-Neural\stable-audio\ - Launch: Choose "Local Model" option
✅ Benefits: Completely offline, no GPU server, no Python installation, instant generation
The TensorFlow Lite models have some quality trade-offs:
- Timing issues: Generated rhythms may not be perfectly quantized
- Quality reduction: TFLite quantization affects audio precision
- High RAM usage: Expect significant memory consumption during generation
- Fixed duration: Audio generation locked to 10 seconds (audiogen/TFLite limitation)
- No STEMS separation: DEMUCS stem separation not available in local mode
For live performance and production use, the server-based approach still provides better quality, timing precision, variable duration, and STEMS separation.
Get FREE API access - No setup required!
- Contact me for a free API key (only 10 slots available)
- Download VST3 from Releases
- Contact: [email protected]
Complete control with local server - for developers and advanced users!
- Python 3.10+ installed from python.org
- Get Stability AI access: Request access to Stable Audio Open on Hugging Face
- Wait for approval: Usually takes a few minutes to a few hours
🪟 Windows:
# Download the Python installer
# From releases: OBSIDIAN-Neural-Installer.py
python OBSIDIAN-Neural-Installer.py
# Alternative: Clone and build
git clone https://github.com/innermost47/ai-dj.git
cd ai-dj
python installer.py🍎 macOS:
# Download from releases
# OBSIDIAN-Neural-Installer-macOS.dmg (recommended)
# OR OBSIDIAN-Neural-Installer-macOS.pkg (system-wide)
# Alternative: Use the executable
chmod +x OBSIDIAN-Neural-Installer-macOS
./OBSIDIAN-Neural-Installer-macOS🐧 Linux:
# Download from releases
chmod +x OBSIDIAN-Neural-Installer-Linux
./OBSIDIAN-Neural-Installer-Linux
# Alternative: Build from source
git clone https://github.com/innermost47/ai-dj.git
cd ai-dj
python installer.pyThe installer automatically handles:
- Virtual environment creation
- All Python dependencies (PyTorch, FastAPI, Stable Audio Tools, etc.)
- AI model downloads (Gemma-3-4B, 2.49 GB)
- CUDA/ROCm detection and installation (if GPU available)
- VST3 compilation (optional, requires build tools)
- Configuration setup
The installation creates a GUI control panel with:
Server Interface Features:
- System Tray Support: Minimize to tray with green triangle icon
- 🚀 Server Control Tab: Real-time status, start/stop/restart buttons
- ⚙️ Configuration Tab:
- API Keys management with credit system (🔓 UNLIMITED, ✅ 50/50 credits)
- Secure Hugging Face token storage with built-in verification
- Model settings and server configuration
- 📝 Logs Tab: Real-time server output with color-coded messages
- First-Time Setup Wizard: Guided configuration for new installations
- Launch the server interface:
python server_interface.py - First launch: Setup wizard will guide you through configuration
- Hugging Face Token: Enter your approved token (built-in verification available)
- API Keys: Generate keys with credit limits or unlimited access
- Start the AI server from the GUI interface
- API Keys prompt:
- "Use stored API keys for authentication?"
- Yes: Use API authentication (for production/network access)
- No: Development bypass - no auth needed (for localhost)
- Server URL: Will be displayed (usually
http://localhost:8000)
- Download VST3 from Releases or build from source
- Load OBSIDIAN-Neural VST3 in your DAW
- Server URL: Paste from server GUI (e.g.,
http://localhost:8000) - API Key: Copy from server interface (if using authentication)
Troubleshooting:
- Windows users: Python installer is the simplest approach
- No Hugging Face access? You must be approved for Stable Audio Open first
- Build errors? Download pre-compiled VST3 from releases instead
- API confusion? Choose "No" for simple localhost setup
- Can't connect? Make sure server is running before configuring VST
Pre-compiled VST3 plugins available for all platforms:
- Windows:
OBSIDIAN-Neural-Windows-VST3.zip - macOS:
OBSIDIAN-Neural-macOS-VST3.zip - Linux:
OBSIDIAN-Neural-Linux-VST3.tar.gz
Installation locations:
- Windows:
C:\Program Files\Common Files\VST3\ - macOS:
~/Library/Audio/Plug-Ins/VST3/or/Library/Audio/Plug-Ins/VST3/ - Linux:
~/.vst3/or/usr/lib/vst3/
## 🎯 What OBSIDIAN-Neural Actually Does
Think of it as having an AI jam partner directly in your DAW.
- Type simple keywords → AI generates musical loops instantly
- Real-time generation → No stopping your creative flow
- 8-track sampler with multi-page system → Each track has 4 pages (A/B/C/D) for instant variation switching
- MIDI triggering → Play AI samples from your keyboard (C3-B3)
- Perfect DAW sync → Everything locks to your project tempo with intelligent time-stretching
- Beat repeat & retrigger → Built-in performance effects for live jamming
- Stems separation → Generate with isolated drums, bass, vocals, etc.
- Sample bank → Drag & drop your generated samples between tracks
Example workflow:
- Type "dark techno kick" → AI generates a techno kick loop on page A
- Switch to page B → Type "acid bassline" → Generate 303-style bass
- Use keyboard shortcuts (Q-W-E-R) to switch between pages instantly
- Trigger both with MIDI keys while jamming on hardware synths
- Apply beat repeat for live performance effects
It's like having a TB-303, but instead of tweaking knobs, you describe what you want in plain English. Plus you get 4 variations per track for endless creative possibilities.
What if recording studios could create AI models based on their own recordings?
- Studios develop their unique AI signatures from decades of work
- Musicians access these signatures for their projects
- Revenue sharing creates new income streams for studios and session musicians
- AI becomes a collaborative tool with traceable musical heritage
The concept: Instead of generic AI trained on anonymous data, musicians could work with AI that carries the DNA of legendary studios like Abbey Road, or their local favorite recording space.
Current AI music generation requires expensive hardware that most musicians don't have.
A potential solution: Community-owned GPU network where:
- GPU owners share idle computing power
- Musicians get affordable access to AI generation
- Revenue gets distributed fairly among contributors
- Infrastructure stays independent from big tech
Looking for volunteers to help develop this distributed approach. If you have GPU resources or infrastructure expertise, let's discuss.
OBSIDIAN-Neural explores AI as a creative partner rather than a replacement for human musicians. The goal is building tools that enhance musical creativity while ensuring fair compensation for everyone who contributes to training these systems.
Get involved:
- Studios: Interested in exploring AI signature models?
- GPU owners: Want to contribute to a community network?
- Musicians: Have thoughts on ethical AI music tools?
Contact: [email protected]
📰 Press coverage moved to PRESS.md
Found issues? Create GitHub Issue
Include: DAW name/version, OS, steps to reproduce, expected vs actual behavior
🚀 Local models: Available now (with some limitations)
✅ Server option: Still the best for live performance
OBSIDIAN-Neural is dual-licensed to provide maximum flexibility for different use cases:
Mozilla Public License 2.0 (MPL-2.0) - Free for open source projects and commercial use. If you modify OBSIDIAN-Neural files, those modifications must remain MPL-2.0, but you can combine them with proprietary code in separate files.
For organizations preferring proprietary licensing terms or requiring additional rights beyond MPL-2.0, commercial licenses are available.
Contact for commercial licensing: [email protected]
- Open Source Community: Full access to innovation with file-level copyleft protection
- Commercial Users: Flexibility for proprietary integration while protecting core innovations
- Project Sustainability: Commercial licenses fund continued development
- License Compatibility: MPL-2.0 is compatible with GPL, LGPL, and AGPL projects
Attribution: Please keep attribution to InnerMost47 in all cases.
OBSIDIAN-Neural is provided "AS IS" without warranty of any kind.
By using this software, you acknowledge that:
- You install at your own risk
- You are responsible for choosing appropriate installation paths
- The developer is not liable for any system damage or data loss
- You should have backups before installation
- You understand this is experimental software
Installation Safety: Always review installation paths carefully.
Avoid system directories like /Applications, /System, C:\Windows, etc.
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
