I’ve spent the last few months building toward this, and I’m glad to finally share it. Live-performance plugin OBSIDIAN Neural has entered beta for a new Local Edition, letting musicians generate AI audio loops directly on their own computer. No GPU, no internet, no cloud dependency once it’s installed.
Why Local AI Music Generation Matters
OBSIDIAN Neural has so far depended on a distributed GPU network or a cloud subscription for its real time generation. I wanted to bring local AI music generation to any laptop, so the Local Edition takes a different path.
It runs Stable Audio 3 Medium locally, producing usable loops in roughly 10 to 11 seconds on a typical laptop CPU. That’s quick enough, I think, to keep up during a live set without breaking the momentum.
- Fully offline once set up, nothing to generate requires an internet connection
- Everything stays on the musician’s machine, no data sent anywhere
- No subscription, no credit system to track
- Works across up to 3 machines per activation
- Runs on Windows, macOS (Apple Silicon only), and Linux
Musicians who want access to the other 8 AI engines can still switch the Local Edition into server mode, which requires either a cloud subscription or a self-hosted server. Full plan details are available on the pricing page.
Who Local AI Music Generation Is For
Running AI generation locally isn’t just a novelty to me, it’s the whole point. It solves real problems for specific kinds of musicians. DJs and live performers who can’t rely on venue Wi-Fi get a generation engine that keeps working with zero signal.
Bedroom producers on a budget avoid a recurring subscription entirely, paying once and using the plugin indefinitely. Touring musicians who work on planes, trains, or in basements with no signal get the same generation quality without waiting on a round trip to a server.
And if you’re the type who worries about where your creative material ends up, I built this with a pretty clear guarantee in mind: nothing generated locally is ever uploaded anywhere.
A Performance Tool First
Unlike text to song generators such as Suno or Udio, I designed OBSIDIAN Neural around live playability from day one.
It’s an 8 track sampler with MIDI triggering, quantized page switching, real time crossfading, and a full effects chain (filtering, EQ, compression, distortion, delay, reverb) meant to be manipulated on stage, not just rendered offline.
Producer Moteka called it “the first [AI tool] that actually feels like a real production tool rather than a novelty,” citing its tight DAW integration. Brian Bullock, Founder/Director at RETHINK Studios, put the Local Edition through its paces and said it’s “fast enough that it never breaks momentum in the DAW.” Hearing that kind of feedback is honestly why I keep building this.
Getting Into the Local AI Music Generation Beta
I kept the process simple on purpose, no application, no gatekeeping. Access is free and immediate. Musicians head to the Local Edition page, and the beta promo code is applied automatically during checkout, so there’s no card details and nothing to type in. The whole process, from clicking through to having a working key, takes less than a minute.
Setup itself is straightforward: download the installer for your platform, activate your key inside the plugin, and let it download the Stable Audio 3 Medium model once over an internet connection.
From that point on, every generation happens locally, with no further downloads or connectivity required, even mid set, mid tour, or in a studio with no signal at all.
About OBSIDIAN Neural
I develop OBSIDIAN Neural under the name InnerMost47. The core plugin has racked up over 5,500 downloads and 226+ GitHub stars so far. The Local Edition is a separate, closed source add on layered on top of the open source plugin, currently free during the beta.
I’ve been fortunate to get press coverage in 8 countries, and I presented the project at the AES AIMLA 2025 conference in London and the 160th AES Convention in Copenhagen (May 2026).
More details at obsidian-neural.com.
Electronic/metal music producer and developer, focused on jamming and improvisation. I build my own plugins using the JUCE framework, and I’m currently working on OBSIDIAN Neural and BeatCrafter.
