JZL - 🎵 MiniMax Music3 提示词预设

"JZL - 🎵 MiniMax Music3 提示词预设" (Music3 Caption Preset) is a music-theory form that writes a prompt for you. MiniMax Music3's official caption format is a "structured caption": three stacked sections - Global Metadata (genre/BPM/key/mood/emotion arc/scene/production), Vocal Details (arrangement/range/timbre/delivery/harmony/effects), and Arrangement (the per-section timeline). Getting that format right by hand is a slog. This node turns ~20 dropdowns into that three-part caption, in the shape the model was actually trained to read.

It's aimed at the same "漫剧" (short drama) pipeline as the rest of the pack - you caption a track, feed it to MiniMax's Music3 node, and score your H3 clips with something that matches the mood you set in the video nodes. The style library is dense: 40 presets covering Lo-fi Hip-Hop, Chillhop, Jazz, Bossa Nova, Soul, R&B, Funk, Hip-Hop, Electronic, Ambient, Synthwave, House and more, each with hand-written global/vocal/arrangement blocks in the density of the official reference examples.

How it works

Mostly it's template assembly with taste. The 音乐风格 (music style) dropdown carries the per-style three-part base text; the other dropdowns fill in the structured fields - 风格融合 adds a secondary genre as influences (the official genre-router's dual-style fusion), 速度 picks a BPM band, 调式/调号/拍号 handle key/mode/time signature, 情绪氛围 and 情绪演变 set mood and its arc, 使用场景 covers listening context (studying, driving, gaming…), 段落结构 chooses the arrangement shape, 主奏乐器/辅助乐器 name the lead and support instruments, 制作质感 sets the production feel (clean digital, warm analog, tape lo-fi, vinyl crackle…), and the vocal block covers 人声配置 (instrumental/female/male/choir/acappella), 声部音区, 人声音色, 人声唱法, 和声伴唱, and 人声效果. Flipping 高级参数 reveals the full set; most fields default to "不指定/跟随风格" so you can set two or three and let the style carry the rest.

Output: caption - the STRING you feed to the Music3 caption input.

The input that matters

• 音乐风格 - the one dropdown that does most of the work. Pick this first; everything else is refinement.
• 高级参数 - toggles the advanced fields (style fusion, time signature, emotion arc, structure, etc.).

One honest note from the tooltip: this node writes the caption; Music3's own cfg_scale (default 1.5) and top_k (default 50) live on the official node's advanced parameters, and this node doesn't touch them.

How to install it

Part of the JZL-MiniMax-H3 pack - ComfyUI Manager, search "ComfyUI-JZL-MiniMax-H3", or:

cd ComfyUI/custom_nodes
git clone https://github.com/wjluoxiao/ComfyUI-JZL-MiniMax-H3

Restart. No dependencies - it's pure string building, and it doesn't load or run Music3 itself (that's the official MiniMax music nodes, fed by your caption).

Common issues

Don't over-specify. The defaults are "unspecified/follow style" for a reason - set a genre, a mood, maybe a BPM band, and let the style's arrangement text do its job; stacking every dropdown fights the model's sense of the genre. Also, this is a Chinese-first pack, so the dropdown labels are bilingual but lead with Chinese - if a label looks familiar-but-not-quite, match by the English half.