HiAPI
minimax-music-3 模型详情页 Hero 图:21:9 宽幅,右侧为半透明玻璃声波雕塑与丝带光轨,左侧留亮部空白,奶油色与青蓝、珊瑚粉渐变交融。

MiniMax

MiniMax Music 3

minimax-music-3

Just input lyrics and a music description to generate a stereo song up to about 5 minutes long, complete with vocals and full song structure—ideal for quickly producing song demos and custom soundtracks.

minimax-music-3

Try minimax-music-3

Input

Lyrics

Output

audio
Fill in the parameters and run — results show up here.

Pricing

minimax-music-3 per second

Unit Price $0.0026

Positioning Overview

minimax-music-3 is a text-to-music model released by MiniMax, available through the unified HiAPI.ai interface. By inputting lyrics and a music description, it generates a fully formed song with vocal performance and complete arrangement, supporting section structures such as intro, verse, chorus, bridge, and outro, with each generation controllable to within a few minutes. It is suitable for developers and content teams that need to produce song demos in bulk, create custom background music, or turn existing lyrics into audio.

Key Highlights

  • Dual input control via lyrics and description — lyrics are annotated line by line with section tags, while the description text can specify genre, BPM, key, and vocal style, so structure and style are controlled separately.
  • Complete song form rather than a music snippet — the generated result includes a systematic section arrangement from intro to outro, paired with vocal performance, and can serve directly as a prototype of a finished song.
  • Layered architecture balances structure and vocal detail — the global layer handles long-range song-form progression, while the local layer handles frame-level acoustic detail, making section transitions in long audio more stable.
  • Duration upper bound settable in seconds — the duration parameter supports 1 to 300 seconds, and the model may finish early, making it easy to control output according to the length of your material.

Capabilities

Structured song-form generation: automatically arranges intro, verse, pre-chorus, chorus, bridge, instrumental section, and outro; writing tags per line in the lyrics specifies section assignment, and unannotated lines are discarded, so lyrics need to be formatted in advance.

Conditional control of musical attributes: the prompt can describe genre, BPM, key, emotional progression, instrumentation, and vocal details; the more specific the description, the more precise the style control, making it suitable for batch creation that requires a stable stylistic identity.

Duration upper-bound setting: duration sets the upper limit of generation length in seconds, defaulting to 60 seconds with a range of 1 to 300 seconds; the model may end early, making it convenient to match output to the length of a video or podcast segment.

Reproducible generation: a seed parameter is provided, and the same input with a fixed seed can reproduce the same result; if omitted, a random seed is used, which is suitable for comparing different descriptions or for version management.

Fixed audio output specification: output is 32 kHz, 16-bit stereo WAV, which can go directly into editing or mixing workflows without additional transcoding.

Use Cases

  • Musicians quickly producing song demos — input complete lyrics and a style description to get a prototype of a finished song with vocals and arrangement, used to show direction to a team or client.
  • Custom scoring for video and game teams — generate background music based on BPM and emotional description, set the duration to match the footage length, and place it directly into editing or game audio tracks.
  • Turning existing lyrics into songs — organize written lyrics with section tags, input them, and pair them with a style description to turn them into a listenable vocal version.
  • Batch exploration by genre and mood — use a fixed seed with multiple description sets to compare generation results across genres and keys, quickly screening usable material.

Limitations and Selection Advice

Fidelity to non-Western music styles is limited, and strongly stylized prompts may deviate from expectations; the model card does not state multilingual singing capability, so non-English lyrics should be validated on a short sample first. A single generation is capped at about 5 minutes, and output is fixed as 32 kHz, 16-bit stereo WAV with no other sample rate or encoding options. If you need longer complete works or strict multilingual singing, validate the effect on a short segment first before deciding whether to adopt it.

  • New model New model: minimax-music-3

Related Models

View all models →

Same Type