Free AI Vocal Remover
Split any song into instrumental and vocals for karaoke, covers or remixes — the UVR MDX-Net model runs in your browser. Free, no sign-up, nothing uploaded.
Need it as a video? Put the instrumental on a cover image with the MP3 to MP4 converter, or change the vocal with the voice changer.
Best-quality mode uses UVR-MDX-NET Inst HQ 4 by the Ultimate Vocal Remover team (MIT). Fast and 4-stem modes use Spleeter by Deezer (MIT), via ONNX exports from sherpa-onnx and Best-Practice (Apache-2.0). Keepframe is not affiliated with these projects.
Keepframe's vocal remover separates a song (MP3, WAV, M4A, FLAC up to 10 minutes) into an instrumental karaoke track and an isolated vocal (acapella), downloaded as 44.1 kHz WAV files. Best-quality mode runs UVR-MDX-NET Inst HQ 4, the open model from the Ultimate Vocal Remover project, directly in your browser; Fast and 4-stem modes use Deezer's Spleeter. It's free with no sign-up, no watermark and no upload. In our test a 3-minute song took about 30 seconds in Chrome on a recent Mac.
Facts checked October 10, 2026.
What you get
| Price | Free. No credits, no sign-up, no watermark, no daily limit. |
|---|---|
| Models | Best quality: UVR-MDX-NET Inst HQ 4 (the model behind the Ultimate Vocal Remover desktop app). Fast and 4 stems: Deezer Spleeter. |
| Stems | Vocals + instrumental (karaoke track), or vocals, drums, bass and other. |
| Output | WAV, 44.1 kHz stereo, same length as your song. |
| Privacy | The song is decoded and separated on your device; nothing is uploaded. Models (59 MB best quality, 39–79 MB Spleeter) download once from our CDN. |
| Speed | Measured on a 3-minute song in Chrome on a recent Mac: about 30 s best quality (WebGPU), 15–20 s fast or 4 stems. Without WebGPU, best quality takes roughly as long as the song. |
| Not the right tool for | Phones and older laptops on long songs (memory), or separating guitar and piano — 4 stems puts them in "other". |
How it works
1. Choose a song
Any MP3, WAV, M4A or FLAC up to 10 minutes. It's decoded to 44.1 kHz stereo in your browser.
2. Pick a mode
Best quality for karaoke and acapellas, Fast for a quick preview, 4 stems for drums and bass.
3. Listen and download
Play each stem, then download WAV files. Close the tab and everything is gone.
Keepframe vs other vocal removers
Honest comparison as of October 10, 2026. Online services run bigger models on servers; this page runs open models on your device.
| Keepframe | Online services (vocalremover.org, LALAL.AI, Moises) | UVR desktop app | |
|---|---|---|---|
| Upload needed | No | Yes | No |
| Account / limits | None | Free tiers cap minutes or files; paid plans for more | None |
| Install | None (browser) | None | Download and install (Windows, macOS, Linux) |
| Best model | MDX-Net Inst HQ 4 | Proprietary or larger models (often cleaner) | Many models incl. Demucs and ensembles (cleanest) |
| Stems | 2 or 4 | 2 to 10+ | 2 to 6 |
FAQ
How do I remove vocals from a song for free?
Choose the song, keep Best quality selected and click Separate. When it finishes, download Instrumental for a karaoke track or Vocals for the acapella. No account, no credits.
Is my song uploaded to a server?
No. The song is decoded and separated on your device. Only the AI model downloads once (59 MB) from our CDN; your audio never leaves the browser.
Which AI model does it use?
Best quality uses UVR-MDX-NET Inst HQ 4 by the Ultimate Vocal Remover team (MIT license) — the same model you can pick in the UVR desktop app. Fast and 4-stem modes use Deezer's Spleeter (MIT).
How good is the result compared with paid tools?
MDX-Net Inst HQ 4 gives clean instrumentals on most pop, rock and hip-hop songs. Heavily reverbed vocals or backing choirs can leave faint traces; paid services and UVR ensembles running bigger models on servers or a GPU can be cleaner.
How long does it take?
In our test in Chrome on a recent Mac, a 3-minute song took about 30 seconds in Best quality (WebGPU) and 15–20 seconds in Fast or 4-stem mode. Without WebGPU, Best quality runs on the CPU and takes roughly as long as the song — Fast mode stays quick.
Can I make a karaoke version or an acapella?
Yes. Instrumental is the karaoke track (everything except the voice); Vocals is the acapella. Both are WAV files the same length as the original.
Can I split drums and bass too?
Yes — choose 4 stems, or open the Stem Splitter page. It returns vocals, drums, bass and other.
Can I use the separated tracks commercially?
The tool and models are free to use, but the song's copyright still belongs to its owners. Use separated stems of songs you own or have permission for, or for personal practice.