AI-powered audio processing platform LALAL.AI has announced a major expansion of its VST plugin: multi-stem separation for six distinct audio sources, running entirely offline on the user’s local machine.
With this release, professionals can now split a mixed track into vocals and instrumentals, bass, drums, acoustic guitar, electric guitar, and piano, without leaving their DAW, without uploading files to the cloud, and without an internet connection. The specificity of stems matters: A generic “guitar” or “other” bucket is a starting point; acoustic guitar and electric guitar as separate, clean outputs are a production asset. For sound designers, remixers, post-production engineers, working with existing recordings, the difference between a broad separation and an instrument-level split determines how much manual work follows.
For producers working inside Ableton, FL Studio, Reaper, or any other VST3-compatible DAW, stem separation has historically meant breaking workflow. LALAL.AI’s expanded VST eliminates that loop entirely. Stem separation now happens where the work already is — inside the session and at production speed.
Architecture Behind LALAL.AI VST
The plugin is powered by Lyra, a model developed specifically for local deployment. Lyra is engineered to run efficiently on standard production hardware, delivering separation quality on par with LALAL.AI’s previous flagship generations, which means clean, artifact-reduced stems that hold up in professional contexts without requiring manual cleanup. The trade-off is intentional: Lyra prioritizes speed and accessibility over maximum fidelity, making it the right tool for iteration inside a session rather than final export.
“Local stem separation is the future of audio production. With the launch of our six-stem VST, we are proving that the caliber of our algorithms can now live entirely on the user’s machine. We specifically engineered the Lyra model to provide professionals with the perfect equilibrium between processing speed and isolation precision,” says Nik Pogorsky, Product Owner & Co-Founder, LALAL.AI. “This release reflects the mission of making sophisticated AI invisible and seamlessly integrated into the daily lives of engineers and producers.”
The result is a different kind of workflow: unlimited experimentation, directly in the DAW, at nearly zero latency, with no usage caps and no dependency on network availability.
The offline architecture adds a second layer of value. Professional studios routinely work with unreleased material, such as demos, pre-release albums, client sessions where uploading audio to a third-party server is a legal and confidentiality risk. Local processing eliminates that risk by design.
This combination, granular separation of six stems, local execution, and zero cloud dependency, is currently unique to LALAL.AI’s VST among tools that function natively within a DAW environment.
VST Already in Use
LALAL.AI’s VST is already being used in demanding professional contexts. Sound designers working in automotive UX, where precise deconstruction of layered audio references is a core part of building signature vehicle soundscapes, have integrated the plugin into their production pipeline. The ability to isolate specific instruments without leaving the DAW, and without sending sensitive audio off-device, has been central to that workflow.
Availability
The expanded VST plugin is available now to LALAL.AI’s premium subscribers. It supports VST3 on Windows, macOS, and Linux, with AU (beta) available for macOS. Tested with Ableton Live, Reaper, FL Studio, and other VST3-compatible hosts.
More information: LALAL.AI
