LALAL.AI Tutorial: Separate Vocals & Stems
Step-by-step LALAL.AI tutorial: separate vocals, drums, bass, guitars & more with the Andromeda engine, best settings, and what the free 10 minutes give you.
Read Article →LALAL.AI Review 2026: 10 Free Minutes, No Downloads
Key Takeaways
- The free plan gives 10 minutes and one-minute previews of all 11 stem types, but nothing downloads until you pay
- A 3:10 song cost me about 3 minutes 10 seconds per stem type, and three full splits left 0:29 of the free 10 minutes
- Each full split finished in well under a minute on the free account, with the stems playable in the browser
- Lite is €8.99 a month or €6.75 billed annually; the one-time Master pack is €100 for 750 minutes, €50 on sale in October 2026
- Guitars is the newest stem, and Voice Changer and Voice Cloner now sit beside the splitter
LALAL.AI is a browser-based stem splitter: you drop in a song and it pulls out the vocals, drums, bass, piano, guitars, synth, strings or wind.
For this review I used the free plan with one song. I previewed every stem type, ran three full splits, and then couldn’t download any of them. The cheapest way to keep your stems is Lite, at €8.99 a month or €6.75 a month billed annually.
Great fit for: musicians making remixes or covers, karaoke fans, DJs pulling out acapellas, podcasters who want background music gone, and creators who need a clean vocal or instrumental track.
LALAL.AI 4.6 From €6.75/mo
Try LALAL.AI Free What is LALAL.AI?
LALAL.AI is a browser-based AI stem splitter run by the Swiss company OmniSale GmbH. You upload a song or video, pick one of 11 stem types, and get that stem plus the mix without it. The free plan gives you 10 minutes and previews, paid plans start at €6.75 a month, and the same account now carries a Voice Changer and a Voice Cloner.
How Does LALAL.AI Work?
You pick a stem type, upload the file, listen to a free one-minute preview, then run the full split. The split costs the file’s length in minutes for each stem type, plays back in the browser, and downloads once you’re on Lite, Pro or a pack. There’s no queue to choose; paid accounts start in Fast and fall back to Relaxed.
Using the Web Interface
1
Select Your Stem Type
Choose what you want to extract from your audio
There are 11 options: Vocal and Instrumental, Drums, Bass, Voice and Noise, Guitars, Electric Guitar, Acoustic Guitar, Piano, Synthesizer, Strings and Wind. One type runs at a time, and you always get two stems back: the one you picked and the mix without it.
2
Upload Your File
Drag and drop audio or video files
It takes MP3, OGG, WAV, FLAC, AIFF, AAC, M4A, AVI, MP4, MKV and MOV, up to 20 files at once. The limit is 200 MB per file on the free plan and 2 GB on paid plans.
3
Preview Every Stem for Free
Listen before you spend any minutes
The preview is one minute long and costs nothing. You can preview every stem type on the same upload before you spend a single minute on a full split.
4
Split in Full, Then Download on a Paid Plan
Run the whole track and save the stems
The full split takes the file’s length off your balance for each stem type. It plays back in the browser, but downloads only unlock on Lite, Pro or a pack.
Desktop & Mobile Apps
-
Desktop Apps (Windows/macOS/Linux)
Native apps with batch processing; Pro adds local processing with Lyra and the VST plugin for your DAW
-
Mobile Apps (iOS/Android)
Process audio on the go, upload from device or cloud storage
Info:
Three splitter engines, one default: Andromeda handles vocals, drums, bass, piano and guitars. Perseus and Orion stay in Settings for vocal and instrumental splits, and Lynx is the separate voice network behind Voice and Noise.
What Happened When I Split a Song on the Free Plan
I uploaded a 3-minute-10-second indie pop song with vocals, a track I’d generated on Suno earlier this year. It was an MP3 of about 3 MB, and I left Vocal and Instrumental selected. The previews came back in roughly half a minute.
A preview is one minute long and free, and I could preview every stem type on the same upload without spending any minutes.
Then I ran three full splits: Vocal and Instrumental, Drums and Guitars. Each took well under a minute on the free account on a Saturday morning, and each deducted about 3 minutes 10 seconds. The balance read 0:29 when I was done, so the three splits used 9 minutes 31 seconds of my 10.
The Guitars stem came back almost empty on this track, a few short bursts in a flat line, and the Without Guitars version looked like the full mix. Fair enough: the song has very little guitar in it, and LALAL.AI didn’t invent any.
The finished splits play in full in the browser, but trying to download on a free account just opens the plan picker. The stems stay in my history for two weeks, and I could add another stem type to the same upload without uploading it again.
That was one song, three stem types, on the free plan. It’s a test, not a benchmark.
When I listened back, the Guitars stem held one guitar line, which is about all the song has, and the Without Guitars track sounded like the normal song to me. That’s the result I’d expect from a track with this little guitar in it, and I’d rather hear that than a guitar LALAL.AI made up.
Here’s the same 30 seconds three ways: the original, the Guitars stem and the mix without it. The free plan doesn’t download, so the two LALAL.AI clips are recordings of the browser player.
The guitars clip wouldn’t have shown you much, so I ran a second, shorter track through the vocal split: a 31-second K-pop song with Korean lines that I’d also made on Suno. A song that short fits inside the free one-minute preview, so these stems cost no minutes at all. The free plan doesn’t download, so the two LALAL.AI clips are recordings of the browser player.
Source The original K-pop song, first 30 seconds: First Day, First Day (Suno v6-mini)
0:00 --:--
LALAL.AI Vocal stem from the free preview, recorded from the browser player, same 30 seconds
0:00 --:--
LALAL.AI Instrumental from the same preview: the karaoke version
0:00 --:--
What Features Does LALAL.AI Offer?
The core is the splitter with its 11 stem types and three engines, backed by Enhanced Processing, De-Echo and noise reduction. Around it sit Voice Cleaner for speech recordings, Voice Changer and Voice Cloner for new vocals, batch processing for up to 20 files, and apps for desktop and mobile.
-
11 Stem Types
Vocal and Instrumental, Drums, Bass, Voice and Noise, Piano, Synthesizer, Strings, Wind, and three guitar choices (Guitars, Electric, Acoustic)
-
Advanced Processing
Clear Cut for minimal bleed, Deep Extraction for maximum detail, plus de-echo and noise reduction
-
Voice Cleaner
The Voice and Noise stem as its own tool: strip background music from interviews, clean up podcasts, reduce ambient noise
-
Batch Processing
Process up to 20 files simultaneously for efficient album or project workflows
-
Format Flexibility
Input MP3, WAV, FLAC, MP4, and more. Video support included. Outputs match input quality
-
Voice Changer & Cloner
Apply an artist-style voice pack to a song or recording, or train your own from clean recordings (1 slot on Lite, 3 on Pro)
-
Multiple AI Engines
v6 Andromeda by default, v5 Perseus and v4 Orion for vocal splits, Lynx for voice cleanup
-
Cross-Platform
Web, desktop (Windows, macOS, Linux), and mobile apps (iOS/Android). Pro runs splits locally with Lyra in the desktop app and includes the VST plugin and API
How Much Does LALAL.AI Cost?
The free Starter plan has 10 minutes, previews and no downloads. Lite costs €8.99 a month or €6.75 a month billed annually, and Pro €17.99 or €13.50 the same way. One-time packs start at €100 for 750 Fast Queue minutes, with the Master pack on sale for €50 in October 2026.
| Plan | Annual (Best Value) | Monthly |
|---|---|---|
| Free (Starter) | Annual €0 | Monthly €0 |
| ||
| Lite | Annual €6.75/mo billed annually | Monthly €8.99/mo |
| ||
| Recommended Pro | Annual €13.50/mo billed annually | Monthly €17.99/mo |
| ||
Success:
Best Value: Lite at €6.75 a month on the annual plan covers most people: unlimited Relaxed Queue splits plus 90 Fast minutes. Pro only earns its price if you want the VST plugin, the API or local processing.
One-Time Top-Ups
Prefer paying once? The packs are one-time fees with downloads, batch upload and API access included:
| Pack | Minutes | Price | Queue |
|---|---|---|---|
| Master | 750 min | €100 (€50 on sale in October 2026) | Fast |
| Premium | 3,000 min | €190 | Fast |
| Enterprise | 5,000 min | €300 | Fast |
Info:
Queue System: you don’t pick a queue. Paid processing starts in Fast until the month’s minutes are used, then continues in Relaxed, which is unlimited on Lite and Pro and waits for server capacity. Quality is the same in both.
Pick Your LALAL.AI Plan
Lite at €6.75 a month on the annual plan gets you unlimited Relaxed Queue splits, 90 Fast minutes and downloads. Preview on the free plan first.
Start Splitting with LALAL.AI →Pros and Cons
Pros
- On my track, the Guitars stem held the one guitar line and invented nothing
- 11 stem options, including one Guitars stem plus separate electric and acoustic guitar splits
- One-minute previews of every stem type cost nothing
- Full splits finished in under a minute in my test
- Works with video files, not just audio
- De-echo, noise reduction and lead/backing vocal separation included
- Desktop and mobile apps, plus a VST plugin and API on Pro
- Voice Changer and Voice Cloner on the same account
Cons
- Nothing downloads on the free plan, not even a 10-second clip
- Fast Queue minutes reset monthly and don't roll over
- Every stem type costs the full file length again
- Subscriptions aren't refundable once they start
- Dense mixes and compressed sources can still leave artifacts
- No editing or practice tools, which Moises bundles in
Who Should Use LALAL.AI?
LALAL.AI suits anyone who needs a clean stem out of a finished mix: singers who want an instrumental, DJs after an acapella, podcasters cleaning up speech, and students isolating a part to learn it. If you only remove vocals now and then, a free tool covers you.
Perfect For:
-
Musicians & Remixers
Extract instrumentals to sing over, isolate vocals for remixes, get stems without original recordings
-
DJs & Producers
Create mashups, extract acapellas for transitions, isolate elements for sampling
-
Karaoke Enthusiasts
Generate karaoke versions of any song instantly with clean vocal removal
-
Podcasters & Creators
Remove background music from interviews, clean up audio with Voice Cleaner
Also great for music transcribers isolating instruments to hear complex parts, and students practicing along with isolated bass, drums, or guitar tracks. For a step-by-step walkthrough, see the LALAL.AI tutorial on isolating vocals and instruments.
Not Ideal For:
| Use Case | Why LALAL.AI Isn't the Best Fit |
|---|---|
| Trying it without paying | You can preview and play full splits, but you can't download anything |
| Perfect artifact-free results | AI separation has limits on complex mixes |
| Audio editing needs | It separates stems and changes voices; editing happens in another app |
| Massive catalog processing | Minute-based pricing adds up for large volumes |
| Live processing | It works on finished files, not live input |
How Fast Is LALAL.AI and What Affects Quality?
In my test, previews came back in about half a minute, and each of the three full splits finished well under a minute on a free account on a Saturday morning. Paid plans get the Fast queue first, and I didn’t test those.
Quality depends mostly on the source. Lossless files or 320 kbps MP3s with clearly defined parts separate best. Dense mixes, heavy compression, live room sound and extreme stereo effects can leave artifacts. In Settings, Clear Cut gives you less bleed between stems and Deep Extraction pulls out more detail.
The output matches the input: WAV in, WAV out, an MP3 keeps its bitrate, and a video file gives you the audio stems back.
How Does LALAL.AI Compare to Moises, UVR & Other Tools?
Against Moises, UVR, StemSplit and iZotope RX, LALAL.AI is the simplest paid route to a clean stem: 11 types, free previews, minute-based pricing. Moises adds practice tools and more stem types, UVR is free but needs a capable computer, and iZotope RX is a full restoration suite.
| Feature | LALAL.AI | Moises | UVR (Ultimate Vocal Remover) | StemSplit | iZotope RX |
|---|---|---|---|---|---|
| Stem Options | 11 stem types | Up to 27 stem types | 6 stems (Demucs) | 4–6 stems | 5+ stems |
| Pricing Model | From €6.75/mo | Free tier, paid plans | Free (open-source) | Pay-per-song | $400+ one-time |
| Ease of Use | ★★★★★ | ★★★★★ | ★★☆☆☆ | ★★★★☆ | ★★★☆☆ |
| Batch Processing | Yes | Pro only | Yes | No | Yes |
| Mobile App | iOS & Android | iOS & Android | No (desktop only) | No (web only) | No |
| Best For | Clean stems by the minute | Musicians practicing | Technical users, free | Occasional pay-per-song | Mastering studios |
Bottom Line: LALAL.AI is the straightforward choice when the split itself is the job: pick a stem type, preview it free, pay by the minute. Moises is the better pick for musicians who practice along, with speed control, pitch shifting, chord detection and more stem types on its paid plans. UVR costs nothing but needs a desktop with a capable GPU and some patience. StemSplit works for occasional one-off separations with pay-per-song pricing. iZotope RX remains the studio standard for restoration beyond separation. For the free browser-based option, see my Vocal Remover review and the LALAL.AI vs Vocal Remover comparison.
Info:
Quality Tip: For best results, upload the highest quality source file available. Lossless formats (WAV, FLAC) produce cleaner separations than compressed formats (MP3).
Warning:
Copyright Notice: LALAL.AI grants you rights to use processed audio, but copyright of the original content remains with its owners. Extracting stems doesn’t transfer composition rights. Use separated content in compliance with copyright law.
Tips for Getting the Best Results
1
Use High-Quality Sources
Start with the best audio quality available
- WAV or FLAC files produce the cleanest separations
- 320kbps MP3 is acceptable for most uses
- Lower bitrates create more artifacts in output
2
Preview Before Processing
Evaluate quality before spending minutes
- The one-minute preview is free, so check the chorus and the busiest section before you spend minutes
- Preview a second stem type on the same upload if the first split sounds off
- Compare Andromeda with Perseus on vocal splits
3
Know Which Engine Does What
Different engines work better for different material
- v6 Andromeda: default for vocals, drums, bass, piano and guitars
- v5 Perseus: the alternative for vocal and instrumental splits
- v4 Orion: older results, occasionally useful on older recordings
4
Use Enhanced Processing Wisely
Choose the right mode for your needs
- Clear Cut: less bleed between stems, slightly thinner sound
- Deep Extraction: more detail, more chance of bleed
- Enable De-Echo when the source has reverb, and try Lead/Back Separation on tracks with backing vocals
5
Optimize for Long Files
Manage minutes efficiently
- Split very long recordings into sections
- Upload WAV or FLAC when you can; the stems come back in the same format
- Batch process similar files together
FAQ
Is LALAL.AI free?
Partly. The free Starter plan gives you 10 minutes, and no card is needed to sign up. You get one-minute previews of every stem type, and finished splits play in the browser, but nothing downloads until you pay. In my test, three full splits of a 3:10 song used 9:31 of the 10 minutes.
How much does LALAL.AI cost?
Lite is €8.99 a month, or €6.75 a month billed annually (€81 a year). Pro is €17.99 a month, or €13.50 a month billed annually (€162 a year). If you'd rather not subscribe, one-time packs cost €100 for 750 minutes (€50 on sale in October 2026), €190 for 3,000 minutes and €300 for 5,000 minutes.
How are minutes calculated?
Minutes equal the file's length multiplied by the number of stem types you run. A 5-minute song split three ways costs 15 minutes. Previews are free, so you can check every stem type before spending anything. Fast Queue minutes reset each month and don't roll over.
What's the difference between Fast and Relaxed queues?
Quality is the same in both, the difference is the wait. You can't pick one. Processing starts in the Fast queue until your monthly Fast minutes are gone, then continues in Relaxed, which waits for server capacity. On paid plans, Relaxed has no minute cap. Lite includes 90 Fast minutes a month and Pro includes 250.
Is LALAL.AI safe to use?
Yes. LALAL.AI is run by OmniSale GmbH, a Swiss company in Risch-Rotkreuz, and uploads aren't used for AI training or shared with third parties. Your uploads and stems stay in your account history, with processed stems available there for two weeks. You sign in with an email code, Google or Facebook, and the free plan asks for no card.
How do I cancel a LALAL.AI subscription?
Cancelling takes four steps: hover the minute balance at the top right, open Profile, choose Manage Subscription, then Cancel Subscription. You keep the paid features and your remaining minutes until the end of the billing period. Subscriptions aren't refundable, so cancel before the renewal date if you're done.
What is the LALAL.AI Voice Changer?
Voice Changer lets you pick a voice pack, an artist-style voice, and apply it to a song or recording, with Accent, Tonality and De-echo controls. Voice Cloner is the companion: upload up to five clean recordings (10 to 50 minutes is recommended) to train your own voice pack. Packs sit in slots, with 1 on Lite and 3 on Pro, and you use them in Voice Changer.
Does LALAL.AI work with video files?
Yes. LALAL.AI accepts MP4, MKV, MOV and AVI files directly and gives you the audio stems back. That's handy for pulling dialogue or music out of a video clip. Audio formats work too: MP3, OGG, WAV, FLAC, AIFF, AAC and M4A, with up to 20 files at once.
How does LALAL.AI compare to Moises?
They solve different problems. LALAL.AI gives you 11 stem types, free one-minute previews and a choice of engines, which suits you when the separation itself is the goal. Moises is built for musicians who practice along, with speed control, pitch shifting, chord detection and up to 27 stem types on its paid plans. Moises also has a free tier with a limited number of uploads a month.
How does LALAL.AI compare to Ultimate Vocal Remover (UVR)?
They suit different people. UVR is free and runs locally, but it needs a capable GPU and some setup, and it offers 6 stems. LALAL.AI runs in the browser with 11 stem options and free one-minute previews, but you pay for downloads. If you're comfortable installing software, UVR costs nothing to try.
Do LALAL.AI minutes expire?
Fast Queue minutes do. They reset every month and don't roll over, so unused ones are lost. Relaxed Queue minutes are unlimited on paid plans, so there's nothing to expire there. The top-up packs are one-time fees rather than subscriptions, with an optional Auto Payments setting that rebuys the pack when your balance drops under 150 minutes.
Is LALAL.AI worth the money?
It's worth it if you want piano, guitars, strings or wind on their own, or process 50 or more minutes a month. Lite covers most people. For occasional vocal removal, free tools like Ultimate Vocal Remover or the free Vocal Remover are better value, and the free plan's previews let you check a result before paying anything.
Final Verdict
4.6
LALAL.AI is mainly a stem separator, and it does that well. The free previews are the best part: you can hear what each stem type will give you before you spend a minute, and that’s how I’d start with any song. The catch is that nothing downloads until you pay, so treat the free plan as a trial. The pricing is fair for what you get, with Lite at €6.75 a month on the annual plan.
Strengths: An honest split on my test track, 11 stem types including Guitars, free one-minute previews, splits in under a minute, apps everywhere, Voice Changer and Cloner on the same account.
Weaknesses: No downloads on the free plan, Fast minutes don’t roll over, every stem type costs the full file length, no refunds.
LALAL.AI Shines For:
- Creating karaoke or instrumental versions of songs
- Extracting stems for remixes and mashups
- Isolating instruments for practice or transcription
- Cleaning up audio for podcasts and videos
- Sampling and music production workflows
Consider Alternatives If:
- You need only occasional, quick vocal removal (try free Vocal Remover)
- You have technical skills and prefer free/open-source (try Ultimate Vocal Remover)
- You need full audio restoration, not just separation (try iZotope RX)
Further Reading
- SigSep: open music separation models - Open-Unmix and the research community’s separation campaigns
- Music Information Retrieval: Source Separation - Academic resources on audio analysis and source separation
