暂无中文版本——显示英文。

`POST /v1/music/generations`

Generate music from a text prompt, with optional lyrics. Routes to providers running a music service (ACE-Step). The response is the raw audio bytes (audio/mpeg by default) — not JSON.

Request

curl https://api.inference.club/v1/music/generations \
  -H "Authorization: Bearer $INFERENCE_CLUB_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ace-step",
    "prompt": "dreamy lo-fi hip-hop with mellow piano and soft vinyl crackle",
    "lyrics": "[verse]\nNeon rain on the window...",
    "audio_duration": 60
  }' \
  --output song.mp3

Field	Type	Notes
`model`	string	Required. A model with `service_type: music` from `GET /v1/models`.
`prompt`	string	Required. Style, mood, instruments.
`lyrics`	string	Optional. Use `[verse]` / `[chorus]` section tags.
`audio_duration`	number	Seconds, 5–300. Omit to let the model decide.
`audio_format`	string	`mp3` (default), `wav`, `flac`, `opus`, `aac`.
`inference_steps`	integer	1–200, default 8.
`guidance_scale`	number	0–30, default 7.
`seed`	integer	Fixed seed; or `use_random_seed: true` (default).
`bpm`	integer	Optional tempo hint.
`key_scale`	string	Optional, e.g. `"C major"`.
`vocal_language`	string	Optional, e.g. `"en"`.

Also accepts the visibility and collection sharing fields.

Response

The finished audio, streamed back with Content-Disposition: inline. The request and its audio are stored in your dashboard (and in Dashboard → Music) for replay, sharing, and playlists.

Errors use the standard error format: 404 no_provider when no online provider serves the model, 413 when prompt + lyrics are too long.