Lyra: Local Stem Separation in the LALAL.AI Desktop App and VST Plugin

No uploads. No minutes deducted. No internet required. Lyra is LALAL.AI’s new stem separation model that runs entirely on your hardware. Split tracks into 7 stems offline.

Lyra: Local Stem Separation in the LALAL.AI Desktop App and VST Plugin

We’re introducing Lyra, a new stem separation model that runs locally on your hardware. Unlike cloud-based processing, it doesn’t upload audio to our servers, doesn’t require a stable internet connection, and doesn’t deduct from your processing minutes. Everything happens on your machine, inside the desktop app or your DAW.

What Lyra Is

Lyra is a stem separation model designed to run locally, without sending audio to our servers. Processing happens on your machine, which means it works offline and doesn’t use your processing minutes regardless of how many times you run it.

Right now, Lyra supports seven stems: Vocals, Instrumental, Drums, Bass, Piano, Acoustic Guitar, and Electric Guitar.

Lyra is available in both the LALAL.AI desktop app and the VST plugin. In the desktop app, it works as an alternative to cloud processing. In the VST plugin, it runs inside your DAW, so you can separate stems as part of your regular production workflow without switching between tools.

When It Makes Sense to Use It

Lyra works best when local processing takes priority over maximum separation quality. 

The most common case is early-stage work: experimenting with settings, comparing outputs, or processing a large number of files. It’s also the right choice when your connection is slow or unreliable, or when you simply don’t want to upload audio every time you need a stem. 

For unreleased tracks or NDA-covered material, keeping files on-machine is often a hard requirement, not just a preference.

In the VST plugin, local processing also makes Lyra usable in live situations like DJ sets and on-stage workflows where depending on a network connection isn’t an option.

How It Compares to Cloud Processing

Lyra is a compact model built to run efficiently on consumer hardware, which means it makes different trade-offs than our cloud models. Andromeda, Perseus, and Orion run on high-performance servers and are built entirely around separation quality, with no constraints on computational cost.

For most work, Lyra’s output is good. The gap becomes noticeable when the source material is complex or dense, or when the output has to be at its absolute best: final mixes, pre-release mastering, anything heading straight to a release. Those are the cases where a cloud model is worth the extra step.

A practical approach is to treat them as two stages of the same workflow. Use Lyra to move fast: experiment, compare versions, narrow down what you actually want. Then, when you’re ready to commit to a final result, run it through a cloud model.

Lyra vs Cloud Models
Lyra (plugin & desktop app) Cloud models (Andromeda, Perseus, Orion)
Processing Local, on your device Remote, on LALAL.AI servers
Privacy Full; audio never leaves your machine Audio uploaded to servers
Speed Seconds, no waiting Depends on internet speed & queue
Usage limits None Minutes deducted per file
Offline use Yes No
Stems available Vocals, instrumental, drums, piano, bass, acoustic guitar, electric guitar Wider selection including synth, strings, wind instruments
Quality Excellent for most tasks Maximum, best for final releases
Best for Experiments, practice, batch processing, confidential material Final mastering, complex audio, release preparation

Access and Licensing

Lyra requires an active Pro subscription.

Desktop App

Once activated, the app works fully offline. An internet connection is needed occasionally for a subscription check, but the check happens infrequently and in the background.

VST Plugin

The license is tied to the machine and can be activated on up to three workstations, which covers a typical setup of a studio desktop, a laptop, and a backup machine. Once activated, the plugin works without a network connection. The occasional subscription check is the only exception. 


Follow LALAL.AI on Instagram, Facebook, Twitter, TikTok, Reddit, and YouTube to keep up with all our updates and special offers.

Cookies

For magic to happen, we use cookies. Read our Privacy Policy to learn more.