Remove the Background Music — or Remove the Vocals
Drop a song or a video and let AI pull the voice apart from the music, then keep whichever side you need: a clean vocal acapella, or the instrumental with the vocals lifted out for karaoke and remixes. It runs on Mel-Band RoFormer — the leaderboard-topping family of source-separation models — and takes about a minute.
Free account · Runs on our GPUs · Results in about a minute
Split your track — vocals or instrumental
Sign in to use the Vocal Extractor
This one runs on our GPUs, so it needs a free ClipCraft account. Every plan — including the free tier — comes with monthly tokens you can spend on it.
What the Vocal Extractor can do
Isolate vocals with AI
AI source separation pulls the singing away from the instruments and backing track, so you get a clean acapella instead of a muddy, phased karaoke-filter mess.
Or keep the instrumental
Point the same run the other way and keep the music instead — the full backing track with the voice lifted out, ready for karaoke, a cover, or a remix. Same run, same price, you just pick the side you want.
Three output modes
Vocals only for a bone-dry acapella, Instrumental for the vocals-removed backing track, or Vocals + faint backing to keep a touch of the original mix glued around the voice.
Songs or videos in
Drop an MP3, WAV, FLAC or M4A, or a video like MP4, MOV, MKV or WebM — it reads the audio and hands back the isolated vocals as a ready-to-use MP3.
Real progress, real fast
A live progress bar tracks the actual separation job on our GPUs — a typical 3-minute song is done in about a minute, then you can play and download it right on the page.
Built on Mel-Band RoFormer
The separation runs a Mel-Band RoFormer model on GPU — the transformer architecture at the top of today's stem-splitting leaderboards — so vocals come out clean and natural with minimal artifacts.
Fair per-minute pricing
Every account gets monthly tokens — the free plan includes 100. An extraction is 10 tokens per minute of audio, so a 3-minute song is 30 tokens, and roll-over token packs never expire.
How it works
Sign in and drop your track
Vocal separation runs on our GPUs, so you'll need a free ClipCraft account. Then drag in a song or a video — MP3, WAV, FLAC, M4A, MP4, MOV and more.
Pick the side you want
Vocals only for a bone-dry acapella, Instrumental for the karaoke side with the voice lifted out, or vocals with a faint slice of the original mix underneath. Then hit go — a progress bar shows the real job as it runs.
Play and download
In about a minute your track is ready — preview it in the built-in player and download the MP3.
Frequently asked questions
What does the Vocal Extractor do?
It uses AI source separation (a Mel-Band RoFormer model) to split the singing away from the music in any song or video, then hands you the side you asked for as an MP3: a clean vocal acapella, an instrumental with the vocals removed, or the vocals with a faint backing layer around them.
Do I need an account to use it?
Yes. Unlike ClipCraft's browser-only tools, vocal separation is heavy AI compute that runs on our GPUs, so it needs a free ClipCraft account and uses tokens. Every plan — including the free tier — comes with monthly tokens you can spend on it.
How much does an extraction cost?
10 tokens per started minute of audio, so a 3-minute song is about 30 tokens. The free plan includes 100 tokens a month (roughly three songs), paid plans include more, and roll-over token packs never expire. Free, Saver and Creator plans cap each run at 3 minutes; Studio removes the length limit.
Can I extract vocals from a video?
Yes — drop an MP4, MOV, MKV, WebM or other common video and it reads the audio track and returns your chosen side as an MP3, ready to drop back into your edit. That works for the instrumental too, so you can mute a singer out of a clip and keep the music.
Can it make a karaoke or instrumental track?
Yes. Pick Instrumental and you get the music with the vocals removed — the backing track for karaoke, covers, or remixing. It's the same run and the same cost as extracting the vocals; you're just keeping the other half of the split. For both sides of one song, run it twice.
How long does it take and how good is it?
A typical 3-minute song finishes in about a minute on our GPUs (a little longer on a cold start). Quality comes from Mel-Band RoFormer, the transformer architecture leading today's separation benchmarks, so vocals come out clean with minimal artifacts.
Do more with a free ClipCraft account
Save your projects and unlock the full image, video and audio editors plus a whole suite of tools — free to start, no card required.
Create a free account