LALAL.AI is an AI stem-separation and vocal-removal service that splits a song, podcast, or video soundtrack into individual stems - vocals, drums, bass, piano, guitars, synth, strings, and wind instruments - via a browser tool, a VST plugin, or an API. It's built for musicians, DJs, remixers, karaoke-track makers, and post-production teams who need clean stems without a studio session. Pricing is minutes-based rather than a flat subscription: a free Starter tier lets you preview results, and paid Lite and Pro plans, from €6.75/month billed annually, add downloads and faster processing. Check the current pricing page, since displayed currency and rates can vary by region.
Who builds it
LALAL.AI is built by OmniSale GmbH, a Swiss company based in Risch-Rotkreuz that has focused on AI audio separation since releasing its first neural network, Rocknet, in 2020. The core workflow is simple: upload an audio or video file (200MB on the free tier, 2GB on paid plans), pick which stem or stems to isolate, preview the result, and download it in the source file's original format and bitrate. By 2025 the company reported 6.79 million registered users, up from 3.8 million in 2024 - what it called its strongest year of growth to date.
Core features
- Sixth-generation separation engine "Andromeda" (2025), a transformer model LALAL.AI says was trained on roughly four times more data than its predecessor "Perseus," delivering up to 40% faster processing and up to a 10% improvement in SDR (signal-to-distortion ratio) on the vendor's own benchmark.
- Ten selectable stem types - vocals, instrumental, drums, bass, electric guitar, acoustic guitar, piano, synthesizer, string instruments, and wind instruments - plus a dedicated lead/backing-vocal splitter and a "Voice and Noise" cleanup mode.
- Andromeda gained dedicated drum-and-bass separation in 2026, building on milestones that began with an 8-stem splitter in 2021 and a 10-stem splitter (adding strings and wind) in 2022.
- A native VST plugin, bundled with Pro plans, that runs separation locally inside a DAW across seven stems (vocals, instrumental, drums, bass, piano, acoustic and electric guitar).
- A documented v1 API for developers offering multi-stem separation and voice cloning in a single request, batch processing, and uploads up to 10GB for authenticated calls.
- Independent testing by MusicRadar scored LALAL.AI 15 out of 20 across 11 stem-separation tools reviewed, calling its instrument recognition "excellent" and noting it was one of only two tools able to cleanly extract a piano stem.
Pricing
- Starter (free): 10 minutes of processing in the slower "Relaxed" queue, 200MB upload limit, preview only - no downloads.
- Lite: €6.75/month billed annually (€81/year); unlimited Relaxed-queue minutes plus 90 Fast-queue minutes a month, 2GB uploads, downloads and batch processing unlocked.
- Pro: €13.50/month billed annually (€162/year), or €17.99/month billed month-to-month; unlimited Relaxed-queue minutes plus 250 Fast-queue minutes a month, the VST plugin, and API access.
- One-time Fast-queue minute top-ups are sold separately, e.g. 750 minutes for $50 or 3,000 minutes for $190.
- A custom-quote Enterprise plan covers larger minute volumes and API terms. Prices display in EUR on the pages we checked but can vary by region - check the current pricing page for your local rate.
Who it's for
The free Starter plan is enough to judge whether LALAL.AI's separation quality fits your needs, since it lets you preview - though not download - results. Musicians, remixers, and podcast or video editors doing occasional work will likely find Lite's 90 monthly Fast-queue minutes sufficient; audio professionals running regular batch jobs, needing the VST plugin, or building on the API should look at Pro or the custom Enterprise tier instead.