Niches · Music generation

Write and sing a complete song from a prompt or lyrics

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

0companies, full fit
7companies, fit with a limit
16companies carry the tag
27share of the category's work
12first-hand accounts

What counts as done

A song of at least 90 seconds sung from the supplied lyrics and sections in the requested language, returned as a playable file, or a refusal that names its reason, from one job.

Measured, for our own tests: the sung lyrics match the supplied text and sections, and the job finishes within 5 minutes. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: make a song about this, turn these lyrics into a song, a personalised birthday song, a 2-minute song about my coding session

Does the job, with a limit 7

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

ListingWhy
ElevenLabs Music API BB
Eleven Labs Inc.
Sections carry lyric lines and durations up to 600,000 ms, and a copyright refusal returns bad_prompt with a prompt suggestion. The call is synchronous with no job id or idempotency key, and acc-07 found non-English vocals patchy. source
fal music models BB
Features & Labels, Inc.
Mureka 9.5 sings up to 5,000 characters of lyrics with a target duration and returns timed lyric sections; the queue returns an id first, and errors are typed, with optional content-policy categories. Language fidelity and the 5-minute bar aren't stated, and no idempotency key exists. source
Google Lyria D
Google LLC
Lyria 3.5 sings supplied lyrics with section tags in the prompt's language and returns lyrics with the audio. Refusals arrive without a category (acc-15), a timeout can bill a song with no id (acc-28), one SDK path fails (acc-19), and charges show hours later (acc-10). source
Loudly Music API F
Loudly GmbH
POST /api/ai/manta/songs takes lyrics, gender and instrumental=false, but needs Loudly to switch on manta_access (403 otherwise), has no length field, and states no refusal or language behaviour. source
MusicGPT API D
MusicGPT Ltd
Lyrics, model choice and lyric timestamps go in; a task id and per-request is_flagged with a reason string come back. No maximum length or language support is stated, and no idempotency key exists, so a blind retry may bill again. source
Soundverse API E
Soundverse Inc.
generate_song takes up to 3,000 characters of lyrics with gender or a saved voice, returns a job id at once and takes an Idempotency-Key. Song length isn't a parameter or published, and content refusals have no named error type, only a failed status with error_message. source
Suno MCP F
Suno, Inc. (United States, terms governed by Massachusetts law)
The server turns ideas and lyrics into songs saved to the library, but tool schemas, refusal behaviour and job handling are unpublished, and acc-23 had to stop for a human when only library ids came back. source

Doesn't do this job 5

Listings in the category that do something else, with the reason.

ListingWhy
Beatoven.ai API F
Beatoven Private Limited (with subsidiary Beatoven Inc.)
The spec has one prompt field and no lyrics or vocal option; output is an instrumental with bass, chords, melody and percussion stems. source
Mubert API F
Mubert Inc.
Generation takes prompts, images or playlists with no lyrics input; acc-34 found its Mubert adapter can't sing. source
MusicGen on Replicate B
Replicate, LLC
The inputs are a text prompt and an optional melody clip; there's no lyrics input or vocal output. source
SOUNDRAW API F
SOUNDRAW Inc.
Instrumental background music only; the compose parameters have no lyrics or vocals. source
Stable Audio API F
Stability AI Ltd (UK, EU, Switzerland) and Stability AI US Services Corporation (elsewhere)
No lyrics or vocal parameter on any audio route; acc-34 found its Stable Audio adapter can't sing. source

What stops agents finishing

  • HTTP 200 with PROHIBITED_CONTENT and no category, so the agent can't tell what to change (acc-15)
  • a synchronous call that bills a song it can't return after a timeout (acc-28)
  • no Suno API outside the MCP server, and the unofficial web route breaks on captchas and encrypted downloads (acc-13, acc-20, acc-21)
  • the same model fails through one SDK path and works through another (acc-19)
  • charges show hours after the call, so a retry loop overspends unseen (acc-10)

Around it

Before: inference.chat-completion. After: file-storage.upload-file, social-media.publish-post.

Where this comes from

Weight 27 of 100 in Music generation: 12 of 43 links, the largest group: Lyria, ElevenLabs and Suno song attempts on HN, the Google AI forum and GitHub, of which 7 are failure reports (refusals, lost billed jobs, SDK errors, the unofficial Suno route breaking)

The same as data: /api/v1/use-cases/music-generation.full-song-vocals.json, or the MCP tool get_use_case with id music-generation.full-song-vocals. Other jobs in this category: /api/v1/niches?category=music-generation.

Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.