Skip to main content
POST
Audio Music
lyria-3.5, Google’s music model, is the music model served today. Send a style prompt and, optionally, lyrics in any language (Vietnamese is sung in order); you get back one MP3 song with vocals. The request is synchronous: the song comes back in the response body, usually within a few seconds to a minute.
elevenlabs-music has been paused since 2026-09-12, and minimax-music and minimax-music-pro were retired on 2026-10-08 because their upstream no longer serves new accounts. A request naming any of them returns 404 model_not_found.

Request

application/json body. Only the fields below are read; anything else in the body is ignored.
string
required
lyria-3.5.
string
Style, mood, instruments, tempo, and the length you want (“a 30-second jingle”). Max 2000 characters. prompt or lyrics is required.
string
The words to sing, with optional section tags such as [Verse] and [Chorus]. Max 3000 characters. Sent after the prompt.
string
default:"mp3"
mp3 only. Any other value is refused with 400.
Length and structure are steered by the prompt, up to a couple of minutes per song; there is no exact-duration parameter. Character limits are counted after NFC normalization, so Vietnamese and other accented scripts count visible characters.

Response

200 OK with MP3 bytes. Headers: The song carries Google’s SynthID audio watermark.

Pricing

You are charged only when a song comes back. A refused prompt, an upstream error or a timeout costs nothing. Music is not on the free tier: it needs a paid balance.

Errors

Every error body carries error.code, error.type, retryable and doc_url (Errors). Nothing is charged on any of them.

See also