Extend, restyle, cover or split existing audio
No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.
What counts as done
A new track extended from the source, or separate stems (at least vocals and instrumental), returned as files, with the upload accepted under a stated length limit and copyright screen.
Measured, for our own tests: an extension keeps the source's key and tempo, and the stems sum back to the source. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: extend this track to 3 minutes
, make a cover of my demo
, split out the vocals
, remake this in the style of the reference
Does the job, with a limit 9
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.
| Listing | Why |
|---|---|
| ElevenLabs Music API BB Eleven Labs Inc. | Upload (copyright-screened, half charged if it matches), inpainting by section source, audio-reference chunks on v2 and a stem-separation endpoint all exist. No upload length limit or stem price is stated in the references read. source |
| fal music models BB Features & Labels, Inc. | Demucs returns vocals, drums, bass and other (six with htdemucs_6s) as WAV or MP3; Stable Audio audio-to-audio, inpaint and outpainting extend or restyle. Upload length limits, copyright screening and the Demucs price weren't found on the pages read. source |
| Loudly Music API F Loudly GmbH | Stems (4, or 12 for Manta songs behind manta_access_stems) work only on Loudly-generated songs, not uploads; uploads go to remix bundles and mastering. No upload length limit or copyright screen is stated. source |
| Mubert API F Mubert Inc. | edit_track forks a track replacing or deleting instruments and stems, and regenerate_similar varies it, but these work on Mubert's own tracks; upload of the user's audio, extend and separated stem files aren't documented for the API. source |
| MusicGPT API D MusicGPT Ltd | Extend after a timestamp, cover, remix, inpaint and stems (2_stems vocals and instrumental, or 6_stems) are priced endpoints taking a file or YouTube URL. No upload length limit or copyright screen is stated. source |
| MusicGen on Replicate B Replicate, LLC | input_audio with continuation extends a clip from continuation_start to continuation_end, or conditions on its melody, but no stems, no upload length limit, no copyright screen and no statement that key and tempo hold; output is non-commercial. source |
| Soundverse API E Soundverse Inc. | Extend, remix, inpainting, 2, 6 or 12 stems at $0.10, and reference inputs of about 30 seconds up to 15 MB are documented, plus a $0.03 copyright check. Whether uploads are screened automatically and whether extend keeps key and tempo isn't stated. source |
| Stable Audio API F Stability AI Ltd (UK, EU, Switzerland) and Stability AI US Services Corporation (elsewhere) | audio-to-audio takes 6 to 380 seconds of MP3 or WAV (100 MB, no copyrighted uploads) with a strength control, and inpaint replaces a span. No stems, and no statement that key and tempo hold. source |
| Suno MCP F Suno, Inc. (United States, terms governed by Massachusetts law) | Covers, Extends and Mashups from the library and audio uploads are listed, but no stems, no upload length limit and no copyright-screening statement were found. source |
Couldn't check 1
The docs were closed to us, or silent. These carry no mark either way.
| Listing | Why |
|---|---|
| SOUNDRAW API F SOUNDRAW Inc. | No public API reference exists; the blog shows compose only, so whether audio input or API stems exist couldn't be read. source |
Doesn't do this job 2
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Beatoven.ai API F Beatoven Private Limited (with subsidiary Beatoven Inc.) | No audio input: stems come only for Beatoven's own composed tracks, as instrument parts rather than vocals and instrumental of an upload. source |
| Google Lyria D Google LLC | Inputs are text and up to 10 images; no audio input, multi-turn editing or stems, as the guide's limitations state (acc-06 edits after generation). source |
What stops agents finishing
- unclear whether reference input is gated per model (acc-16)
- stems or vocals need an account flag or a custom plan (Loudly, Mubert)
- copyright screening of uploads may refuse the user's own audio (vendor docs, not tested)
Around it
Before: file-storage.read-file, music-generation.full-song-vocals. After: video-generation.compose-with-audio, file-storage.upload-file.
Where this comes from
Weight 9 of 100 in Music generation: 4 of 43 links: Lyria Studio on HN, a Google AI forum request for music-to-music, a GitHub spec for reference-based restyling, and the HN reference-SFX ask
The same as data: /api/v1/use-cases/music-generation.reference-edit.json, or the MCP tool get_use_case with id music-generation.reference-edit. Other jobs in this category: /api/v1/niches?category=music-generation.
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.