meta/demucs
Demucs by Meta - advanced AI model for audio-to-audio. Delivers high-quality results with fast inference, suitable for both creative and production workflows.
Example output — click Run to generate your own
Send your first request
OpenAI-compatible endpoint with unified authentication and usage tracking.
https://api.sandbase.ai/v1/runmeta/demucsInput Schema
6 parameters · 0 required · 6 optional
| Parameter | Type | Required | Description |
|---|---|---|---|
audio | string | Optional | URL of the audio file to separate into stems |
stems | string[] | Optional | Specific stems to extract. If None, extracts all available stems. Available stems depend on model: vocals, drums, bass, other, guitar, piano (for 6s model) · Default: ["vocals","drums","bass","other","guitar","piano"] |
shifts | integer | Optional | Number of random shifts for equivariant stabilization. Higher values improve quality but increase processing time. · Min: 1 · Max: 10 · Default: 1 |
overlap | number | Optional | Overlap between segments (0.0 to 1.0). Higher values may improve quality but increase processing time. · Min: 0 · Max: 1 · Default: 0.25 |
output_format | string | Optional | Output audio format for the separated stems · Options: wav, mp3 · Default: "mp3" wavmp3 |
segment_length | integer | Optional | Length in seconds of each segment for processing. Smaller values use less memory but may reduce quality. Default is model-specific. |
Output Schema
| Field | Type | Description |
|---|---|---|
id | string | Unique identifier for the generation task |
status | string | Task status: pending, running, completed, failed, timeout |
model | string | Model used for the generation |
outputs | array | Array of output items |
outputs[].url | string | URL of the generated artifact |
outputs[].content_type | string | MIME type (e.g. image/png, video/mp4) |
error | object | null | Error details if failed, null on success |
error.type | string | Machine-readable error type code |
error.message | string | Human-readable error description |
Async Workflow
This model uses asynchronous execution. Submit a request and poll for the result.
- Submit — POST to /v1/run, receive an
id - Poll — GET /v1/run/{id} until status is
completed,failed, ortimeout - Retrieve — Read
outputsfrom the completed response
Code Examples
Ready-to-run snippets
# Step 1: Submit
curl -X POST https://api.sandbase.ai/v1/run \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "meta/demucs",
"audio": "https://static.sandbase.ai/examples/meta/demucs/input_audio_0.mp3",
"stems": [
"vocals",
"drums",
"bass",
"other",
"guitar",
"piano"
],
"shifts": 1,
"overlap": 0.25,
"output_format": "mp3",
"prompt": "a beautiful sunset over mountains"
}'
# Step 2: Poll result (replace <id>)
curl https://api.sandbase.ai/v1/run/<id> \
-H "Authorization: Bearer YOUR_API_KEY"API README
Demucs
Demucs is built for music source separation, with this exact route focused on stem-level audio isolation. Demucs by Meta - advanced AI model for audio-to-audio. Delivers high-quality results with fast inference, suitable for both creative and production workflows. It is designed for production workflows that need a clearly scoped media operation instead of a generic assistant response, preserving the supplied direction while returning an artifact ready for review, iteration, or downstream assembly.
The request surface keeps creative direction and delivery choices explicit through stems, shifts, overlap, audio_url, sync_mode, output_format. Teams can lock source material, describe the intended result, choose supported output controls, and rerun the same request structure across variations. This makes experiments easier to compare and larger pipelines easier to operate without hiding important decisions inside an opaque preset.
Highlights
- Stem-Level Audio Isolation. The
meta/demucsroute is dedicated to stem-level audio isolation, keeping inputs and outputs matched to this operation. - Vocals And Accompaniment Workflows. Within
meta/demucs, documented prompt and media controls direct vocals and accompaniment workflows while keeping the brief close to the result. - Time-Aligned Separated Audio. The
meta/demucsresult provides time-aligned separated audio in a machine-consumable form for review and downstream production. - Route-specific control. For
meta/demucs, documented controls includestems,shifts,overlap,audio_url,sync_mode,output_format, defining the supported variation space for this exact route.
Pricing
| Configuration | Price |
|---|---|
| Base-price example | $0.0035 |
| Rate | $0.0007 per second |
When to Use
| Scenario | Why it fits |
|---|---|
| Choose this route | When the deliverable is specifically stem-level audio isolation, not a neighboring generation or editing task. |
| Production iteration | When creative teams need repeatable vocals and accompaniment workflows with an explicit request payload. |
| Automated pipelines | When time-aligned separated audio must flow into review, storage, compositing, or publishing. |
| Controlled variations | When a stable brief should be rerun while changing only documented inputs. |
| Prototype to scale | When the same meta/demucs contract should serve a single experiment and a larger batch later. |
Prompt Guide
Lead with the subject or source, then state the transformation, important spatial or temporal details, style, and finish. Supply only documented fields, keep media URLs reachable, and vary one control at a time when comparing outputs.
{
"stems": [
"vocals",
"drums",
"bass",
"other",
"guitar",
"piano"
],
"shifts": 1
}
Technical Specs
| Property | Value |
|---|---|
| Model ID | meta/demucs |
| Required inputs | None documented |
stems | array; default: ['vocals', 'drums', 'bass', 'other', 'guitar', 'piano'] |
shifts | integer; default: 1 |
overlap | number; default: 0.25 |
audio_url | string |
sync_mode | boolean; default: false |
output_format | string; options: wav, mp3; default: mp3 |
segment_length | integer |
safety_tolerance | string; default: 6 |
Related Models
mirelo/sfx1.6/text-to-audiominimax/music/2.5