We must make aesthetic choices, not AI.
Vibe Together is not interested in prompt-to-song AI technology. We are exploring ways to use technology to aid the player but never replace them in the creative process. The aesthetic decision, the subjective call, and the final shape of a song should stay in human hands. AI can help the artist explore ideas they care about but should not generate a piece. We cannot prompt our creativity away.
The new generation of AI music tools is undoubtedly powerful. Prompt-to-song systems such as Suno can turn a line of text into a complete recording. That is a technical achievement. But it is also a very different creative contract from the one Vibe Together is trying to build. Text goes in, a waveform comes out, and most of the musical decision-making happens inside a black box the player cannot inspect, tune, route, or perform. We believe that this approach takes away something fundamental about what it means to create as a human and is the wrong path. Technology can help but we should use it wisely.
The player owns taste
Our principle is simple: subjective and aesthetic decisions belong to the player. The player decides what feels good, what feels strange, when a loop has enough tension, when a voice should stay rough, and when a mistake has become the best part of the piece. AI can help exploration, but it should not take ownership of the taste decision.
That stance matters because music is not only output. The human creative act of making music, testing a sound, changing your mind, dragging a branch across the field, and hearing the result is the experience. A tool that skips the act may produce a song, but it also removes the part of the process this platform is built around.
White-box instead of black-box
We use some machine learning tools to help players explore synth sounds to try new sound design or help them modify a specific phrase for accessibility reasons. If they want to test a quick articulation change on a set of notes, they can use our Conductor feature to help execute that. It does not silently rewrite the room. The player previews the plan, applies it explicitly, and can undo the result. Players can also make optional raw material that can be trimmed, pitched, routed, and replaced. Future Heritage does not train models on creator content. ElevenLabs describes its music model as trained on licensed stems and music.
Machine help, human reins
We hope our approach offers a better path for how technology can aid the music creation process. Making music should be fun, game-like, and community based. It is the best way to combat the deluge of pure prompt-to-song content that produces little connection for listeners or makers. These new tools should help a player bring an artistic idea to life inside the compositional toolset. They should make the material more playable and more legible. They should not replace the playful judgment of the person making the piece.