MiniMax Music 2.6

Music 2.6 is a high-fidelity generative music model developed by MiniMax, designed to synthesize full-length songs from text descriptions and lyrics. Released in April 2026, it represents a significant update to the MiniMax audio suite, focusing on reduced latency—down to under 20 seconds for initial output—and expanded creative control. The model produces studio-quality audio with sample rates up to 44.1kHz and supports a wide variety of musical genres, ranging from traditional Chinese music to modern electronic styles like House, Trap, and Drum & Bass.

One of the primary advancements in Music 2.6 is the introduction of the AI Cover feature, which allows users to reimagine existing tracks in different styles while preserving the original vocal melody. Additionally, the model includes a Lyrics Optimizer that can automatically generate song lyrics based on a provided style prompt, as well as an Instrumental Mode for creating backing tracks or cinematic scores without vocal components. The model is optimized for improved mid-to-low frequency acoustics, providing more impactful bass and tighter percussion than previous versions.

Key Capabilities

• Song Structure Control: The model supports over 14 specific structural tags, including [Verse], [Chorus], [Bridge], [Drop], and [Outro], enabling users to dictate the emotional arc and arrangement of a composition.
• Technical Precision: Users can specify exact musical parameters such as BPM and key (e.g., "E minor, 120 BPM") within the prompt, with the model demonstrating high reliability in matching these technical specifications.
• Multilingual Support: While capable of generating music in various styles, the model currently offers its strongest vocal performance and pronunciation accuracy in English and Mandarin Chinese.

Prompting and Tips

For optimal results, prompts should combine stylistic keywords with specific technical requirements. A detailed prompt like "90 BPM, acoustic guitar ballad, male vocal, emotional" helps define the tonal foundation. When inputting lyrics, using newline characters to separate phrases and bracketed tags to define sections ensures the model correctly handles the song's pacing and intensity shifts. For instrumental tracks, users should enable the dedicated instrumental toggle or use [Inst] tags to suppress vocal generation.