AI music generator for covers: what Sume cannot do, and a swap

Can an AI music generator cover an existing song on Sume? No. Sume's Music Router takes a prompt and one optional image. It has no audio input to feed a song into, and nothing in its docs offers a song-to-song or voice-swap path. What it can do is write a new track in the same genre, tempo and mood as the song you have in mind, with lyrics you supply.

That difference is the whole point of this post. A cover reuses someone else's composition, and the rights questions are theirs to answer. A new track in a similar style is an original request you can judge on its own. The rest of the page says where covers are heading, what Sume does instead, and how to write that brief.

Where are licensed AI covers heading?

The rights-holder side is moving. As reported on September 10, 2026, Universal Music and ElevenLabs announced a platform, still in development with no release timeline, meant to support AI remixes, mashups, new interpretations of tracks, and personalised vocal experiences using music supplied by participating artists. The same report mentions Udio preparing to launch with support from Merlin and non-Sony majors. Both are described as licensed, opt-in routes, not general-purpose cover tools.

Why does licensing decide this?

The Munich ruling against Suno in July, reported by a law-firm tracker, is a reminder of why licensing is the sticking point. Its summary says the court found that handing protected songs to users through outputs infringed the right of making them available to the public. The judgment is not final, and it concerns Suno, not Sume or Google. It does show why a tool that quietly reproduces a known song is a bigger risk than one that makes something new.

What can I make instead on Sume?

If you wanted a cover for a video, you usually wanted a feel, not a particular recording. You can describe that feel without naming the song. Keep these points:

• Genre and era: "1970s soul ballad, dry and close" instead of an artist's name.
• Tempo and key as numbers, since the docs' brief template uses both.
• Instruments with texture, two to four of them.
• Your own lyrics, if you have them. Lyria 3.5 supports lyrics per Google's page; Sume returns any model-reported lyrics in result.lyrics.
• An arc: a named moment such as a key change at 0:20.

What should the prompt leave out?

Do not paste the original song's title or lyrics. The docs say a policy rejection should be handled by revising the flagged content while keeping the musical brief, and the stored artist-reference guide shows what that looks like.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: soul-ballad-001" \
  -d '{
    "prompt": "1970s soul ballad, 68 BPM, F major. Warm Rhodes, upright bass, brushed drums, a swelling string line at 0:20. Female lead vocal, intimate and close. Lyrics: [Verse] The porch light stays on / for the ones who come home late. A 45-second track."
  }'

What about the singer's voice?

Voice is the other half of a cover, and it is where Sume is also thin for this use. Voice cloning exists in the Sume app but not as a Music Router field, so the music API cannot make a track sung in a particular person's voice. The vocal you get is whatever the engine writes from your description, such as "female lead vocal, intimate and close", and you should check it by ear before it goes anywhere public.

That also means you cannot ask for a likeness of a named singer. A description of vocal character is fine. A named person's voice is outside what the router offers.

How do I get the file and what does it cost?

The result comes back as an audio artifact on media.sume.com once the job completes. Send the same Idempotency-Key on a retry of the submit so a dropped connection does not bill twice. Each accepted generation costs the fixed Music price, and a retake is a new generation.

If the goal is a cover of a specific recording for release, Sume is the wrong tool and a licensed cover platform, or the rights holder, is the right conversation. If the goal is a video that feels like that era or genre, the brief above gets you most of the way, and it is yours to describe.