MiniMax launches music model that creates 5-minute songs

AI company MiniMax released a new music model called MiniMax-Music3 on August 14. It can create complete songs up to 5 minutes long from lyrics and music descriptions. The output is a 32kHz, 16-bit stereo WAV file. The model uses a layered design with a large language model (8B parameters) for song structure and a smaller one (0.6B) for fine audio details. It keeps the melody, rhythm, and singing voice consistent throughout the track, including intro, verses, chorus, and bridge sections. Sample songs and a GitHub page are now available for users to try.