OBSIDIAN-Neural

🎵 Real-time AI music generation VST3 plugin for live performance

⚡ Quick Start: Get your API key and start generating in minutes — no GPU or setup required. 🔥 Free codes available ↓

📄 Late Breaking Paper - AIMLA 2025 - Presented at the first AES International Conference on Artificial Intelligence and Machine Learning for Audio (Queen Mary University London, Sept 8-10, 2025)
🎓 Tutorial - From DAW setup to live performance (French + English subtitles)

Live AI music generation in your DAW

Draw-to-Audio Live Demo

Watch the draw-to-audio feature in action - Live improvisation session showing real-time sketch-to-sound generation:

Raw improvisation showcasing the experimental draw-to-audio workflow: sketch patterns → AI interprets visuals → generates matching loops on the fly.

What Obsidian Neural does

Type words → Get musical loops instantly. No stopping your creative flow.

• 8-track sampler with MIDI triggering (C3-B3)
• 4 pages per track (A/B/C/D) - Switch variations instantly by clicking page buttons
• Draw-to-sound - Sketch your ideas visually and let AI interpret them musically
• Perfect DAW sync - Auto time-stretch to project tempo
• Real-time generation - No pre-recorded samples

Text example: Type "dark techno kick" → AI generates techno loop → Trigger with MIDI while jamming

Drawing example: Draw a chaotic pattern → AI interprets visual elements as musical characteristics → Generates matching loop

What Obsidian Neural does NOT

OBSIDIAN Neural is NOT a song generator like Suno or Udio. It's a performance tool for musicians:

• No full songs in one click - You build your track loop by loop
• No lyrics or vocals - Focus on instrumental loops and sound design
• No auto-arrangement - You're the composer, AI is your loop generator

Think of it as: A creative sampler that generates sounds on-demand, not a "press play and get a song" tool.

Features

Text-to-Audio Generation

Type natural language descriptions and get instant musical loops:

• "dark techno kick"
• "ambient pad with reverb"
• "808 bass with distortion"

Draw-to-Audio Generation

Express your musical ideas visually (original concept by A.D.) - Each track includes a drawing canvas where you can sketch patterns, shapes, or abstract concepts. The AI analyzes your drawing using a Vision Language Model (VLM) and translates visual elements into musical descriptions:

1. Click the Draw button on any track
2. Use pencil, brush, spray, or eraser tools
3. Choose colors and brush sizes
4. Click Generate to send your drawing

How it works:

• Your drawing is encoded as Base64 image data
• A Vision Language Model interprets the visual elements using musical vocabulary
• The generated textual description is sent to the audio generation model
• Result: A unique sound that matches your visual concept

Example workflow:

• Draw sharp, angular lines → AI interprets as "aggressive staccato synth"
• Sketch flowing curves → AI interprets as "smooth flowing pad"
• Use red/orange colors → AI might associate with "warm analog tones"
• Create chaotic patterns → AI interprets as "glitchy percussion"

Canvas features:

• 512x512 drawing area
• 4 brush types (Pencil, Brush, Spray, Eraser)
• 10 color presets
• Adjustable brush size (1-50)
• State preservation per track/page

This feature bridges the gap between visual creativity and sonic experimentation, allowing producers to explore sound design through intuitive visual metaphors.

Note: This draw-to-audio feature is currently experimental and under active development.

Quick Start

🟣 Cloud Inference API (Recommended)

Generate AI loops directly from your DAW with zero setup. Professional quality.

1. Download VST3 from Releases
2. Get your API key from obsidian-neural.com
3. Load the VST in your DAW
4. Configure in the plugin:
   • Click the ⚙️ settings button (top right) or wait for the first-time setup dialog
   • Choose "Server/API" mode
   • Enter Server URL: https://ai-harmony.duckdns.org/obsidian
   • Enter your API key
   • Click "Save & Continue"

👉 Detailed setup guide:
📘 Getting Started
🎚️ First Step - Using the Plugin in Your DAW
🎨 Draw-to-Audio - Express Your Musical Ideas Visually
🎛️ Bank Management - Organize and Manage Your Generated Sounds

Pricing:

• Free: 10 credits (10 samples) - Try it out
• Starter: €14.99/month - 500 credits (500 samples)
• Pro: €29.99/month - 1500 credits (1500 samples)
• Studio: €59.99/month - 4000 credits (4000 samples)

Each generation costs 1 credit (1 LLM generation + 1 audio generation). Samples up to 30 seconds, ~10s generation time.

Benefits:

• Zero setup - Works immediately
• Professional quality
• Variable duration (up to 30s)
• No GPU or powerful hardware required
• Perfect for live performance

🔵 Self-Hosted Server + GPU (Advanced)

Best for privacy, customization, and unlimited generations.

1. Get Stability AI access
2. Follow build from source instructions
3. Run server interface: python server_interface.py
4. Download VST3 from Releases
5. Load the VST in your DAW
6. Configure in the plugin:
   • Click the ⚙️ settings button (top right) or wait for the first-time setup dialog
   • Choose "Server/API" mode
   • Enter your local server URL (default: http://localhost:8000)
   • Leave API key empty for local server
   • Click "Save & Continue"

Benefits: Unlimited generations, full privacy, variable duration

Requirements: GPU with CUDA support, Python environment

🟢 Local Models (Offline)

Runs completely offline. No servers, Python, or GPU needed.

1. Get Stability AI access
2. Download models from innermost47/stable-audio-open-small-tflite
3. Copy to %APPDATA%\OBSIDIAN-Neural\stable-audio\
4. Download VST3 from Releases
5. Choose "Local Model" in plugin

Requirements: 16GB+ RAM, Windows (macOS/Linux coming soon)

Current limitations:

• Fixed 10-second generation
• Some timing/quantization issues
• High RAM usage

Which Option Should I Choose?

Community

🎯 Share your jams! I'm the only one posting OBSIDIAN videos so far. Show me how YOU use it!

📧 Contact: Form
💬 Discussions: GitHub Discussions
📺 Examples: Community Sessions
🌐 Website: obsidian-neural.com

Download

VST3 Plugin:

• Windows
• macOS
• Linux

Install to:

• Windows: C:\Program Files\Common Files\VST3\
• macOS: ~/Library/Audio/Plug-Ins/VST3/
• Linux: ~/.vst3/

Status & Support

🚀 Active development - Updates pushed regularly
⭐ 136+ GitHub stars - Thanks for the support!
🐛 Issues: Report bugs here
📊 API Status: Check status page for real-time service monitoring

🎯 Community Milestone

Road to 200 Stars! Currently at 136+ 🌟

When we hit 200 stars, we're celebrating with a community giveaway:

Prize: 1 year of Starter Pack free access (€179.88 value)
Eligibility: Active community members (stars, discussions, issues, contributions)

👉 Join the discussion

Every star, contribution, and piece of feedback helps make Obsidian Neural better. Thank you for being part of this journey! 🙏

License

• 🆓 GNU Affero General Public License v3.0 (Open source)

AI Model: Stability AI Community License

More Projects

🎵 Music Production Tools

🥁 BeatCrafter - Intelligent MIDI drum pattern generator VST3
Proof of concept for live performance with foot controller. Generates authentic drum patterns that evolve from simple to complex.

🎛️ Randomizer - Generative music studio
Experimental web-based music creation tool with procedural generation.

🎧 Audio Content

🎵 YouTube Channel - Original compositions
Electronic music, ambient soundscapes, metal, and experimental productions by InnerMost (2010-2023).

💬 Connect

📧 Contact: Form
💬 GitHub: @innermost47
🌐 Website: anthony-charretier.fr

Credits & Attribution

Developed by InnerMost47 (Anthony Charretier)

Special thanks to:

• A.D. for the original draw-to-audio concept
• Stability AI for Stable Audio Open
• The open-source community
• All beta testers and early adopters

OBSIDIAN-Neural - Where artificial intelligence meets live music performance.

Made with 🎵 in France