MiniMax Music 2.6

The music generator to pick when the track needs words. Describe a piece and it writes one; hand it lyrics and it performs them, either yours, up to 3,500 characters with \`[Verse]\` and \`[Chorus]\` structure tags, or a set it writes for itself from the prompt. Eleven Music, the other model here, is instrumental only, so anything with a vocal on it comes from this one.

Prompts want genre, instrumentation, tempo and mood, and they want restraint. "Solo piano, sparse, slow, melancholy" produces something usable under a voice-over; "epic cinematic soundtrack" produces a trailer bed that will fight every other element in the cut. Specifying what should *not* be there (no drums, no vocals, no build) is as useful here as it is when prompting a clip model that generates its own audio.

It is also the model that gives you the encoding: sample rate, bitrate and MP3, WAV or PCM are all yours to set, where Eleven Music publishes one fixed MP3. What it does not give you is length: the model decides how long the piece runs, so a track that has to sit under a cut of a known duration is a trim in the edit rather than a parameter here.

The practical use is a bed under a sequence rather than a standalone piece. Generate the track after the cut exists, so the shape of the piece is known, and prompt for the feel the edit needs rather than for a song you would listen to on its own. “join_clips” can lay a generated track under an assembled run at no extra credit cost, which is where most of these end up.