AI Song Splitter
Commercial use OK
567 models
No watermark
No sign-up needed
Model:
+ GPT-5, Claude, Gemini
Splitting track into individual stems...
This may take 30-60 secondsSeparated Stems
Advanced options
Result
Tokens running low.
Get More Tokens
❤️ Love Free.ai? Tell your friends!
Sign up to get a referral link and earn 30,000 tokens per friend.
Want more?
Sign up free: 30,000 tokens/day
Sign Up Free
Processing your request...
Split any song into 4 stems with AI: vocals, drums, bass, other. MP3 / WAV in, ZIP out. Open-source Mel-Band RoFormer + SCNet, fully free.
How to Use AI Song Splitter
1
Enter your input
Type text, upload a file, or describe what you want. No account needed.
2
Click generate
Our AI processes your request in seconds using the best open-source models.
3
Download & share
Download, copy, or share your result. Free for personal and commercial use.
Use this tool via API
Automate this tool from your own code. OpenAI-compatible REST endpoint, Bearer-token auth, no extra SDK required. Token costs match the web interface.
curl -X POST https://api.free.ai/v1/music/generate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"prompt": "upbeat synthwave instrumental", "duration": 30}'AI Song Splitter - FAQ
It separates any song into 4 stems with AI: vocals, drums, bass, and other (synths/piano/guitar/strings). Each stem is its own WAV file. Vocals come from an open-source Mel-Band RoFormer model; drums, bass and other come from an open-source SCNet model.
Vocal remover keeps everything except vocals (instrumental output). Isolate vocals does the opposite (acapella output). Song splitter gives you all four stems separately so you can mix and match. Same separation models under the hood.
A ZIP archive with 4 WAV files: vocals.wav, drums.wav, bass.wav, other.wav. Each is the same length and sample rate as the input, so they layer cleanly when re-mixed in a DAW.
MP3, WAV, M4A, FLAC, OGG, AIFF, and most common audio formats. Max upload 50 MB (about 5 minutes of MP3).
6-15 seconds per minute of input on our A100. A 3-minute song takes about 25-40 seconds.
Excellent on well-mixed studio tracks - vocals are nearly perfect, drums and bass very clean, "other" sometimes has a bit of crosstalk from neighboring stems. Lo-fi recordings, live shows, or heavily-effected tracks produce noisier separations.
Four stems (vocals, drums, bass, other) is the split the open-source models we run are trained for, and it is where their quality is strongest. We do not offer a 6-stem split with separate guitar and piano.
Yes - the result page shows individual download buttons for vocals.wav, drums.wav, bass.wav, other.wav alongside the all-in-one ZIP.
Personal use (practice, study, remixes for non-commercial enjoyment) is generally fair-use OK. Sampling, publishing, or commercially releasing stems requires the song's sync / mechanical / sample license. You're responsible.
About 3,000 tokens per song, flat per-request (regardless of song length, since the separator runs once on the whole track). Free daily-pool tokens cover most use.
Spleeter is an older open-source separator. The Mel-Band RoFormer and SCNet architectures we run are newer and score well above it on the public MUSDB18 benchmark; on 50 MUSDB18 test excerpts our vocal stem measures 11.3 dB SDR. Moises is a paid subscription, while Free.ai is pay-per-use with a free daily pool.
Yes - POST audio to /v1/audio/separate/ with `mode=stems` to get all four stems back. Or specify `mode=vocals` / `mode=instrumental` for the single-stem variants. Docs at /api/.
How would you rate this tool?