JZL - 🎵 MiniMax Music3 提示词预设（双人）

"JZL - 🎵 MiniMax Music3 提示词预设（双人）" is the duet edition of the pack's Music3 caption builder. It has everything the single-voice version has - the 40-style library, the structured Global/Vocal/Arrangement caption format, the mood and production dropdowns - and then it layers on the part that a duet needs: a second, distinct voice.

Getting two voices right in a text prompt is where caption writing normally falls apart. "Two singers" is vague and the model guesses. This node makes the split explicit: a 角色分工 (role division) dropdown that defines the relationship - equal lead, A leads and B harmonizes (or the reverse), A takes verses and B the chorus, B takes verses and A the chorus, or call-and-response - plus a separate voice block for each character covering 角色A/B配置 (female/male/child/androgynous), 音区 (vocal range: soprano through bass), 音色 (timbre), and 唱法 (delivery: falsetto, rap flow, bel canto, vibrato, etc.). The defaults are a sensible pairing - female lead, male counterpart - but every axis is yours to move.

How it works

Same engine as the single-voice node: dropdowns assemble a structured caption in the official Music3 format. The duet-specific bits are the role-division field, the doubled character blocks, and two extras: 纯人声无伴奏 (a true acappella toggle that drops the instrument sections and makes the whole track vocals-only) and the shared 和声伴唱 / 人声效果 fields for how the parts blend (double-tracked harmonies, stacked harmonies, backing vocals; tape delay, warm reverb, auto-tune…). The output caption feeds MiniMax's Music3 node, which runs the actual generation - this node only writes the prompt, and the pack is explicit that Music3's own cfg_scale/top_k advanced parameters stay on the official node.

The inputs that matter

• 角色分工 - the relationship between the two voices. This is the duet's whole point; the rest of the fields are the same refining dropdowns as the single-voice node.
• 角色A配置 / 角色B配置 - each voice's gender/vocal setup.
• 纯人声无伴奏 - acappella mode, for when you want the two voices naked.

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 prompt assembly, and it doesn't run Music3 itself.

Common issues

The duet-specific trap is contradicting yourself: if you set 角色分工 to "A lead, B harmony" but then give both voices "bel canto" and identical ranges, the model has nothing to distinguish them on. Give the two characters genuinely different timbre and range at minimum. And as with the single-voice version, don't over-specify every field - a genre, a mood, a role split, and two distinct voice blocks is usually enough to get a duet that actually sounds like two people.