STEM‑KARAOKE: CURRENT STATE

Goal: karaoke app that splits a song into stems, lets the user re‑map those stems to new instruments, and plays back the result in real‑time. You can sing over it, or play live over the stems, but the app itself does not necessarily record or export the final mixed‑down track.

WHAT IT IS

Real‑time karaoke engine that:

Think: “Enter Sandman on a French horn choir, you sing over it, but the app does not save the full remixed track.”

FEATURES (CURRENT IDEA)

Core modes

  • Single‑player stem‑karaoke: pick a song, pick your instrument mapping, sing.
  • Live‑only playback: no export of the remixed track, no “save” to the app.
  • Stem‑mute / solo modes: duck, solo, or turn off drums, vocals, bass, etc., like a standard karaoke app.

Game modes

  • “Opponent assigns instrumentation”: Player A picks the song; Player B picks the instrument mapping (e.g., “this song will be played on xylophone”); aber B picks the instrument mapping (e.g., “ class="strong">No‑export battle mode: ephemeral, single‑use play, no recorded “remix” saved by the app.
  • Optional logging of “sessions” only as metadata (song, user, instruments), not the audio itself.

WHERE WE ARE LEGALLY

Key question: is this a derivative work or not?

In short: the app is not “clean,” but it is significantly safer if it is live‑only and non‑exporting.

LEGAL RISK (rough, optimistic scale):

APPROACHES TO MAKE IT MORE FEASIBLE

  1. Launch in “live‑only, no‑export” mode.
    • Stems are generated client‑ or server‑side, but the remixed track is never written to disk or offered as a download.
    • Users can record themselves externally (phone, camera) at their own risk.
  2. Work with a licensed catalog.
    • Use standard karaoke‑style licensing for the underlying song (mechanical, possibly sync).
    • Seek explicit permission from labels / artists for “remix‑style re‑mapping of stems” for specific tracks [web:111][web:98][web:109].
  3. Frame it as a “remix‑as‑service” layer for labels.
    • Partner with platforms that provide “remix‑ready” stems (e.g., AudioShake, OpenMusic AI Stem Splitter, etc.) and integrate their licensing flows [web:98][web:113].
    • Charge labels a fee for enabling fan‑remixes, rather than ripping their catalog [web:98].
  4. Keep game modes ephemeral.
    • Opponent‑picks‑instrumentation is fine as long as the app itself does not save the mixed‑down track.
    • Log only metadata: song, user, mode, instruments, timestamps.

WHAT TO AVOID

If you are tempted to do any of those, someone with a law degree probably wants to talk to you first.

SHORT‑TERM ROADMAP (FOR THE COFOUNDER)

  1. MVP: “live‑only” stem‑karaoke.
    • Stem‑separation backend (OpenMusic, LALAL, etc.).
    • Real‑time mapping GUI (mute/solo/remap). No export. No save.
  2. Playability layer.
    • “Opponent assigns instrumentation” game mode, still live‑only.
    • Local‑only recording allowed, but not exported through the app.
  3. If it gains traction, negotiate a licensing spine.
    • Reach out to karaoke / stem‑platform partners (AudioShake, OpenMusic, LALAL, etc.) to explore official remix‑ready catalogs.
    • Build a whitelisted, permissioned track list.
  4. Optional: label‑sanctioned remix‑as‑service.
    • Labels release stems via the app; you take a rev‑share for remixing, exports, social features.
    • Now you are in the “20–30% risk” bucket, assuming proper contracts.

BOTTOM LINE

This idea is not legally clean by default, but it is technically and structurally viable if you:

If you behave like a respectful, label‑adjacent remix layer, the odds of this actually becoming a real product are decent. If you try to be a “pirate‑engine‑first,” odds drop fast.

REFERENCESer B picks the instrument mapping (e.g., “s="unbold"> [web:98] AI‑powered stems and derivative‑work licensing. [web:108][web:111] What counts as a derivative work (remix, edit, etc.). [web:84][web:113] AI stem‑separation karaoke workflows. [web:109] How to remix a song legally (or not).