88Z AI, 14 August — MiniMax has released the open-source music model MiniMax-Music3, capable of generating lyricized songs up to five minutes long. Using a hierarchical architecture with an 8B global model and a 0.6B local model, it outputs 32kHz stereo WAV audio while maintaining consistency in theme, rhythm, and vocal identity across full song structures.
At a glance
- Uses a hierarchical architecture with an 8B global model and a 0.6B local model
- Supports generating 5-minute 32kHz stereo songs while maintaining structural coherence
The update
MiniMax has officially open-sourced its latest music generation model, MiniMax-Music3, which is capable of producing songs with lyrics lasting up to five minutes. Technically, the model employs a hierarchical design strategy: an 8-billion parameter global model handles long-range semantic understanding and overall song structure, while a 0.6-billion parameter local model restores fine-grained acoustic information to ensure audio detail. The final output is a 32kHz stereo WAV file. The model effectively maintains thematic coherence, rhythmic stability, and vocal identity consistency during generation, fully presenting standard song arrangement progressions from intro and verse to chorus.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- IT之家Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.