OBSIDIAN-Neural — AI-Powered Live Music VST3 / AU / Standalone Plugin

Transform AI into a live music instrument with OBSIDIAN-Neural, a neural network plugin designed specifically for live performance and on-the-fly music creation. Available as VST3, AU, and Standalone.

Local Edition — Runs 100% on Your CPU (Beta, Free)

• Stable Audio 3 Medium runs entirely offline on your own CPU - no GPU, no cloud, no account, no subscription.
• ~11s per generation on a recent laptop CPU, alongside a full DAW session.
• Free during the beta period - activate with the beta code available on obsidian-neural.com, no card required.
• Offline forever once set up - activate on up to 3 machines (Windows, macOS Apple Silicon, Linux).
• Can optionally switch to server mode to access the other 8 AI engines, which requires a subscription or your own self-hosted node.

Key Features

• LLM-powered contextual prompt generation that understands musical context (with optional bypass mode for faster generation).
• 9 specialized AI models assignable per track - Stable Audio 3 Medium, Foundation-1, StableBeaT, EDM Elements, and more (server mode only, aside from Stable Audio 3 Medium which also runs locally).
• Distributed GPU provider network - community-hosted nodes serve generations with 85% revenue sharing.
• Stable Audio Open via fal.ai as automatic fallback when no provider node is online.
• Stereo loop generation (~30s) synchronized to your DAW's sample rate and tempo.
• No pre-recorded samples - everything is generated live.

Prompt Bank with Editor

• Dedicated prompt editor with per-model keyword guidance - each of the 8 AI engines has its own keyword library (genres, elements, moods, qualities, negatives) reflecting how that model was trained.
• Format hints per model - natural language for stable-audio-open, tag-based for Foundation-1, structured format for gluten_v1, etc.
• Clickable example prompts tuned for each model.
• Organize prompts by category (Drums, Bass, Lead, Pads, FX, Vocal, etc.).
• Drag-and-drop prompts onto tracks to assign both prompt and AI model in one gesture.

8-Track AI Sampler

• Independent MIDI triggering (C3-B3) for instant loop playback.
• True stereo processing across all tracks with dual VU meters per track.
• 4 pages per track (A/B/C/D) - switch variations instantly, quantized to bar boundaries.
• 8 sequences per page - each with independent MIDI patterns (up to 64 steps).
• Advanced stereo waveform editor with zoom, precise loop points, and drag & drop to DAW.
• Automatic BPM detection (MiniBPM) and time-stretching (Signalsmith Stretch) - locks samples to host tempo with zero pitch drift.
• Individual stereo track outputs for professional mixing workflow.
• 4 pair crossfaders - blend each deck A/B pair independently with model-aware color morphing.
• Per-track Reverse button - flip playback direction instantly, quantized to the next step.
• Transient Scatter - repositions playback to a randomly chosen strong-transient point within a zone every 2 steps, for glitchy, unpredictable rhythmic movement.
• Beat repeat with random retrigger intervals - stutters, rolls and glitch effects on the fly.

Built-In Effects

• Per-page ADSR envelope - Shape the dynamics of each variation independently, editable directly on the waveform.
• Per-track gain control - Adjust each sample's level (-12 / +12 dB) before mixing, with visual waveform feedback.
• Per-track multi-mode filter - LP/HP/BP with 12 or 24 dB slopes, drive, cutoff and resonance for sculpting each voice.
• Per-track 8-band graphic EQ + master EQ - Independent frequency shaping on every voice and global bus, from 40 Hz to 15 kHz.
• Per-track compressor + master compressor - Full dynamic control on each voice and on the master bus, with threshold, ratio, attack, release and makeup gain.
• Per-track limiter + master limiter - Peak protection on every voice and on the master bus with adjustable release and ceiling.
• Per-track distortion - 6 character modes (Soft, Hard, Tube, Fold, Diode, Cubic) with pre/post gain and high-pass cutoff.
• Per-track chorus - Modulation effect with rate, depth, delay, feedback and dry/wet control for added stereo width.
• Tempo-synced delay send - 8 time divisions (1/16 to 2 bars), Stereo / Ping-Pong / Mono modes, per-track send level.
• Reverb send - Per-track reverb with size, damping, width and mix controls.
• Airwindows Console6 master bus - Analog-modeled saturation for cohesive mix glue.

Built-In Step Sequencer

• 16-step programmable patterns per track with multi-measure support.
• Perfect DAW sync with host transport and tempo.
• Live pattern editing during playback.
• Loop marker integration for sequencing specific audio sections.

Standalone Mode

• Run OBSIDIAN Neural without a DAW, with built-in transport and tempo control.
• Ableton Link support - bidirectional network sync of tempo and start/stop with any Link-enabled app.
• Exclusive WASAPI mode on Windows for low-latency audio without ASIO drivers.

Companion Mobile Controller

• Free Android MIDI controller app with plug-and-play default mapping.
• Bidirectional feedback - LED states reflect arm/play status, knob positions follow page changes.
• Auto-configured on first connection - open the MIDI panel and load the default mapping.

Live Performance Ready

• Global custom prompts shared across all projects.
• MIDI Learn system for complete hardware controller mapping, with persistent user mappings.
• Sample bank with category organization - every generation is auto-saved and reusable across projects.
• Session save/load functionality.
• Background processing - works with VST window closed.
• Automatic sample rate detection ensures perfect audio quality.

How to Get Started

• Want it free right now? Grab the Local Edition beta - runs on your CPU, no card needed.
• Prefer the cloud? Generation runs on our distributed GPU network - no GPU or powerful hardware required on your end. Any laptop can run the plugin.
• Try FREE with 20 credits - Generate 20 AI samples to test the plugin.
• Variable duration up to 30 seconds.
• ~30s generation time (cloud) or ~11s (Local Edition).
• Works immediately - VST3, AU, and Standalone formats included.
• Perfect for live performance.

Subscription Plans (Cloud)

• Base: €7.99/month - 150 credits (150 samples).
• Starter: €11.99/month - 300 credits (300 samples).
• Pro: €14.99/month - 500 credits (500 samples).
• Get your API key at obsidian-neural.com.

Become a GPU Provider

If you have a GPU and want to earn from the network, you can host your own provider node and receive 85% of the subscription revenue distributed monthly across active providers. The provider kit and full revenue redistribution history are public on the website.

Perfect For

• Live electronic music performance.
• Creative studio experimentation.
• Bedroom producers seeking AI collaboration.
• DJs and live performers wanting AI generation during their set.
• Standalone live setups (no DAW required).
• Anyone exploring human-AI musical interaction.

Requirements

• Format: VST3, AU, Standalone.
• OS: Windows 10/11, macOS 10.14+, Linux Ubuntu 18.04+.
• RAM: 4GB minimum (cloud mode). Local Edition: 16GB minimum.
• CPU: modern multi-core CPU required for the Local Edition (no GPU needed).
• Local Edition on macOS: Apple Silicon (M1+) only - Intel Macs not supported.
• Cloud mode: internet connection + API key from obsidian-neural.com. Local Edition: internet required only for initial setup.
• For Ableton Link: local network access.

Open Source (Cloud Version)

• License: GNU AGPL v3.0 (Plugin) / Stability AI Community License (AI Model).
• Full source code: GitHub.
• Models: HuggingFace.
• Research Paper: AIMLA 2025 Conference.
• The Local Edition is a separate, closed-source paid build - not covered by this license.

Resources

Website: obsidian-neural.com.

Community: GitHub Discussions.

Developed by InnerMost47 (Anthony Charretier).