Clumi separates vocals, drums, bass, piano, and guitar from songs and videos in the browser, with extra tools for noise cleanup, key and BPM detection, and audio-to-MIDI conversion.
Used this tool? Rate it
Used this tool? Rate it
Clumi (written as Clumi AI on many of its own pages) is an AI vocal remover that runs in the browser and splits a song or a video soundtrack into separate vocal and instrumental tracks. The homepage states the purpose directly: remove vocals from a song and create clean instrumental tracks for karaoke, remixing, and music production.
Two-track separation is only the starting point. The same site hosts a multi-stem splitter, a lead and backing vocal splitter, noise, echo, and reverb removers, a key and BPM finder, and an audio-to-MIDI converter, plus a downloadable desktop program for Windows and Mac.
The pages reviewed do not name a registered company behind the service; the footer simply reads "Copyright © 2026 Clumi." The Terms and Conditions define the "Country" whose laws govern the agreement as Hong Kong SAR China.
Several pages call the tools free with no account needed, yet the site also sells an unlimited subscription and prepaid minute packs that unlock full tracks beyond a 60-second limit.
The two groups follow different limits: separation runs are capped per file and tied to the paid 60-second full-track unlock, whereas key and BPM detection is promoted without usage limits.
The vocal splitter and cleanup tools follow the same upload, preview, and download sequence on their own pages.
These are the vendor's descriptions of intended use, not independent test results.
Long concert recordings or podcast episodes may exceed the web upload caps; the desktop program is described as handling longer files.
It is a weaker fit for some readers:
A giveaway listing for Clumi AI Vocal Remover 2.0.0, written by the vendor and published on a third-party software site in August 2026, describes what the desktop build adds over the web tools:
Paid plans are tied to an account: after payment, the subscription is activated by logging in to the program with the registered account, without a license key.
The pricing page offers two purchase types: an Unlimited Plan billed as a subscription and Minute Packs bought once. Prices below are the USD figures rendered on the pricing page as captured on September 28, 2026.
| Plan | Price shown | What is included |
|---|---|---|
| 1-Month Plan | $12.99, billed monthly as a subscription | Unlimited processing and exports, full tracks without the 60-second limit, fast queue, MP3 and WAV export |
| 1-Year Plan | $6.66/mo billed annually, shown beside $12.99 and a "50% OFF" label; the annual charge is not displayed | Same as the monthly plan |
| 10 minutes | $9.99 ($0.99/minute), one-time purchase | Full tracks without the 60-second limit, fast queue, MP3 and WAV export |
| 300 minutes | $49.99 ($0.17/minute), one-time purchase | Same as 10 minutes |
Minute Packs are pay once, use anytime: minutes are deducted based on the duration of the uploaded file, and the purchase never expires. The Unlimited Plan instead covers any number of files during the subscription period.
When captured, an "Anniversary Deal" banner offered 20% off all plans with a coupon code, though the code itself was not shown in the page text. Students can get an extra 50% off all plans, and the offer also extends to faculty, staff, and teachers who contact support from an academic email address.
The free tier is described inconsistently:
Against that, the site header displayed a "3 File Left" counter when captured, and both paid options advertise unlocking full tracks as a benefit. None of the pages reviewed spells out how many free files are allowed per period or exactly how the 60-second limit applies to free results.
Refund terms differ between the pricing page and the standalone Refund Policy:
Cancellation runs through support: the Refund Policy asks subscribers to contact the company by email to cancel, and says a reminder email goes out 7 days before an auto-renewal. Access continues until the end of the billing period after auto-renewal is cancelled.
Clumi was not among the nine tools in a July 2026 hands-on stem-separation test by a music-technology publication, so there is no side-by-side quality result for it. That test, which discloses affiliate links, describes these commonly compared alternatives:
| Tool | How it is offered | Free option | Paid model |
|---|---|---|---|
| LALAL.AI | Web, desktop and mobile apps, plus a real-time VST/AU plugin | Several files free | Subscriptions or one-off fees |
| Moises | Web application plus desktop and mobile apps | Free Starter plan with a monthly separation cap | Plans with unlimited separations and higher quality |
| Gaudio Studio | Website, with a mobile app version | Free minutes of audio with MP3 output | Minute bundles that add a fast-track service and WAV files |
In the same test, LALAL.AI was named the standout for transient punch and consistency on drums.
Clumi's minute packs resemble Gaudio Studio's bundles, while its Unlimited Plan is closer to Moises's unlimited tiers. No DAW plugin appears on the Clumi pages reviewed, whereas Moises offers a Stems plugin on its Pro Plan and LALAL.AI has its real-time plugin.
Clumi's statements about uploaded audio come from marketing copy, not from the Privacy Policy.
The Privacy Policy text reviewed does not mention uploaded audio files, how long uploads are kept, or whether they are used to train AI models.
Partly. The web tools work without an account, but paid plans advertise unlocking full tracks beyond a 60-second limit, and the header shows a remaining-file counter.
Yes. A dedicated lead and backing vocal splitter puts the main vocal and the background harmonies on separate tracks.
Under the Refund Policy, Minutes Pack purchases are final and non-refundable, although a partial refund for unused minutes may be considered in special circumstances. The pricing page FAQ does not mention this exclusion.
No. The site states the vocal remover is not available for iOS now; iPhone users can use the browser version.
