Stem splitting API

Live

Split a song into vocals and instrumental, or vocals, drums, bass and other.

From $0.01794 per 1-min song, +$0.00299 per extra minute with Demucs

Demucs

Input

Song URLaudio_urlA public https link to the song.

6603e279-4bfc-4077-b73b-63ddbc87585a_example.mp3Reading the song's length…

Split intostems

Songs under a minute cost at least $0.01.

Example

Example outputMade with this prompt

0:00 / 0:00

Vocals
Drums
Bass
Other

Compare Stem splitting API models

Every model behind this endpoint, with live prices. Try one in the playground or open its page.

DemucsMeta$0.01794 per 1-min song, +$0.00299 per extra minute
Vocals + instrumental · Vocals, drums, bass, other
Page

Every model of this tool with its price
		Official			Lengths / sizes	Badges	Actions
DemucsMeta	$0.01794 per 1-min song, +$0.00299 per extra minute	No official price	—	Vocals + instrumental · Vocals, drums, bass, other	15 min		Page

Call Stem splitting API from your code

One endpoint for every model above. Change the model field to switch.

POSThttps://api.opennan.com/v1/audio/separations

GET/v1/models?category=musicEvery model of this endpoint, with prices

Get an API key
Create a key in the console. It uses the same balance as the playground.
Create a key

Send your first request

The same request the playground sends. Set OPENNAN_KEY to your key.

curl https://api.opennan.com/v1/audio/separations \
  -H "Authorization: Bearer $OPENNAN_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"demucs","audio_url":"https://example.com/song.mp3","stems":"vocals"}'

# Then poll until "status" is "succeeded" or "failed":
curl https://api.opennan.com/v1/tasks/<task id> -H "Authorization: Bearer $OPENNAN_KEY"

Poll the task
The call answers right away with a task id. Ask for the task until it finishes:
GEThttps://api.opennan.com/v1/tasks/{id}
queued
Waiting for a worker
running
Being made
succeeded
Done: outputs[].url holds the files
failed
Failed: you are not charged

Parameters

Built from the live catalog: every value listed here is one the API accepts.

Parameters
Name	Type	Allowed values	Default	Effect on price
`model`requiredWhich model runs the call.	string	demucs	—	—
`audio_url`requiredThe song to split (https link).	URL	—	—	Billed by song length
`stems`optionalWhich parts to split the song into.	string	vocals · all	vocals	—
`callback_url`optionalWe send the finished task to this URL.	URL	—	—	—

Errors and how to fix them

401Missing or wrong key
Send the header Authorization: Bearer with a key from the console.
402Balance too low
Add credit in Billing, then send the same call again.
400A value is not accepted
Read error.message: it names the field. Use a value from the parameter table.
429Too many requests
Wait a few seconds and try again, waiting longer after each 429.

Add Opennan to your AI app

One install gives your AI every Opennan tool: video, images, chat, voice, music, sound effects, 3D, captions and dubbing. Pick your app and follow the steps.

MCP serverhttps://api.opennan.com/v1/mcp

Works in the Codex CLI, the Codex app and the Codex IDE extension: they share ~/.codex/config.toml. If the app or IDE does not see OPENNAN_KEY, use the config.toml option with your key in it. Codex asks before each tool call: approve it to go on.

Get an API key
Create a key in the console. It starts with sk- and uses the same balance as the website.
Create a key

Save your key as OPENNAN_KEY

# macOS / Linux (zsh shown; bash: ~/.bashrc)
echo 'export OPENNAN_KEY="sk-..."' >> ~/.zshrc
source ~/.zshrc

# Windows (PowerShell), then open a new terminal
setx OPENNAN_KEY "sk-..."

3
Install the Opennan plugin: the server plus a skill that teaches your AI to pick models, confirm prices and wait for results
```
codex plugin marketplace add Opennan/opennan-ai-plugin
codex plugin add opennan@opennan
```

Or add only the server

codex mcp add opennan --url https://api.opennan.com/v1/mcp --bearer-token-env-var OPENNAN_KEY

Or add this to ~/.codex/config.toml

[mcp_servers.opennan]
url = "https://api.opennan.com/v1/mcp"
http_headers = { Authorization = "Bearer sk-..." }

4
Check that opennan is listed
```
codex mcp list
```
Restart Codex, then ask for anything
“Make a 5 second video of rain on a neon street, a song to go with it and a poster image.”

Replace sk-... with your own key and keep it private: anyone who has it can spend your balance. Your app can list the tools before you add a key; making anything needs one.

This tool from your AI

Once Opennan is added, your AI calls this tool with arguments like these.

Tool: separate_audio
Example arguments: { "model": "demucs", "audio_url": "https://example.com/song.mp3", "stems": "vocals" }