MiniMax’s New Model Generates 20‑Second Covers in Any Style

The AI music landscape is rapidly evolving, and MiniMax has released a major update that adds a Cover capability and a new Music 2.6 model.

Cover feature : The system can keep the original melody while freely re‑arranging accompaniment, replacing lyrics, and transferring style. Tests show that a popular internet song can be re‑lyricized with workplace‑related text without mismatching syllable density, and a classic folk tune can be turned into a soulful, bar‑room jazz version. The model also converts a high‑density rap diss by Annie Hathaway into a gentle campus‑folk rendition, demonstrating natural‑sounding vocal synthesis and smooth guitar accompaniment.

Music 2.6 improvements : Compared with the previous version, latency is reduced to under 20 seconds, instrument coverage expands to over 100 types, and the human voice becomes more free‑form and realistic. Mid‑low‑frequency optimization yields stronger bass and drum presence, making the model especially suitable for rhythm‑heavy pop, electronic, or rock tracks. Prompting with simple style keywords (e.g., “strong‑beat pop, rock”) can generate a complete song with lyrics in as little as 20 seconds.

Demonstrations : A plain “Happy Birthday” melody is rendered in EDM style; the traditional “Auld Lang Syne” is re‑imagined as soul music; a Jiangsu folk song is transformed into a rock version with distorted guitars and driving bass; and a rap‑style diss is turned into a mellow folk ballad.

Music Skills ecosystem : Three open‑source skill packages are released on GitHub and can be invoked via the mmx-cli tool. minimax-music-gen provides vocal, instrumental, and cover generation modes with strong prompt control. minimax-music-playlist builds personalized playlists by scanning local music libraries (Apple Music, QQ Music, Spotify, NetEase) and generating 3‑7 track lists based on mood, language, and style preferences, supporting real‑time feedback. buddy-sings creates pet‑centric songs by reading a pet’s name and personality from Claude Code and singing from the pet’s perspective.

All skill packages are hosted at their respective GitHub repositories, allowing users to set an API key and run a local AI music studio with bilingual command‑line switches.

Agentization : MiniMax’s roadmap moves toward an agent‑driven workflow where the AI can understand multi‑step user intents, reducing interaction friction and enabling complex creative tasks that were previously impossible without manual intervention.

In conclusion, the technical gap between AI‑generated and human‑performed music is narrowing; Music 2.6 delivers perceptible gains in latency, controllability, and musicality, while the Cover capability opens a new dimension of secondary creation, lowering the barrier for anyone with a melody idea to produce polished, style‑specific tracks.