MiniMax Music
API-first full-song generation; Chinese songs at batch scale, with versions pinned.
Model profile
- Vendor
- MiniMax
- Form
- 服务
- Input
- 歌词 · 文本 · 参考音频
- Output
- 音频
- Licence
- 产出商用、署名与再分发按 MiniMax 开放平台条款；不同模型版本的许可与配额可能不同
- Verified
- 2026-10-31
Related
The API first, the browser second
For developers and batch work, MiniMax Music is primarily an API: one request with lyrics and a music description returns a finished track with vocals. The web app and third-party platforms work too, but batch runs, automation and product integration are smoother through the API. It does not provide a session file or stems; changing one part means regenerating or reaching for a separation tool. For a content team, its value is turning song production into a queueable, retryable, billable job.
Lyrics, description, section tags
Two fields matter most: lyrics and prompt. Lyrics can carry tags such as [Verse], [Chorus] and [Bridge]; the prompt describes genre, mood, voice, instrumentation and production. The response is audio data or a download link, with format, sample rate and duration ceiling set by the model version. The more the prompt reads like a production brief, the steadier the result; "a nice pop song" gets you the average.
| Field | Role | Common mistake |
|---|---|---|
| lyrics | What is sung and how it is split | Lines too long, tags missing |
| prompt | Style, voice, arrangement | Stacked adjectives, no instruments |
| reference audio | Locks timbre or style | Conflicts with the text |
| version | Capability and limits | No pin, so results drift |
Chinese songs at batch scale
Mandarin diction, tones and sentence endings are its main advantage in the domestic market; the same lyric sits closer to how a local singer would phrase it. In batch work, fix lyrics and version, change one prompt variable at a time, and a usable direction appears quickly. For short video, ads and spoken-word beds, drafts come far faster than manual arranging. Current versions state a ceiling around five minutes; actual length and response format belong to the docs.
Responses, retries and version drift
The call is synchronous, so long songs wait; timeouts and rate limits are normal at peak, and retries need backoff rather than full concurrency. Two calls from one request differ in vocal timbre and mix, so stitching across runs sounds like a change of singer. A model upgrade changes how old prompts behave; pin the version in production and archive lyrics, prompt and version together. Download links usually expire, so store the file immediately.
Commercial limits and alternatives
Confirm: commercial use and release, credit requirements, data retention and training use, concurrency and quota, and licence differences between versions. The MiniMax platform docs and terms are the authority. Alternatives: TME Tianqin for stems and post-production; Suno for a browser service; YuE or DiffRhythm for local control.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- MiniMax 开放平台的音乐生成 API 文档（歌词、描述、返回格式与限制）。
- MiniMax Music 模型说明与版本列表（时长、采样率与可用模式）。
- MiniMax 开放平台服务条款（产出商用、数据使用与配额）。