Nodes/ComfyUI-JZL-MiniMax-H3/JZL - 🔗 MiniMax H3 参考总线(解包)
ComfyUI Node

JZL - 🔗 MiniMax H3 参考总线(解包)

Unpack one reference bus back into 48 named ports

By wjluoxiao·Created 2 months ago·Updated 13 days ago· 107
JZL - 🔗 MiniMax H3 参考总线(解包)
  • bus
  • 角色A
  • 角色B
  • 角色C
  • 角色D
  • 角色E
  • 角色F
  • 角色G
  • 角色H
  • 场景A
  • 场景B
  • 场景C
  • 场景D
  • 场景E
  • 场景F
  • 场景G
  • 场景H
  • 道具A
  • 道具B
  • 道具C
  • 道具D
  • 道具E
  • 道具F
  • 道具G
  • 道具H
  • 音频A
  • 音频B
  • 音频C
  • 音频D
  • 音频E
  • 音频F
  • 音频G
  • 音频H
  • 视频A
  • 视频B
  • 视频C
  • 视频D
  • 视频E
  • 视频F
  • 视频G
  • 视频H
  • 视频A(音频)
  • 视频B(音频)
  • 视频C(音频)
  • 视频D(音频)
  • 视频E(音频)
  • 视频F(音频)
  • 视频G(音频)
  • 视频H(音频)

"JZL - 🔗 MiniMax H3 参考总线(解包)" (Reference Bus Unpack) is the receiving end of the pack's reference bus. It takes a single JZL_REF_BUS wire - the bundle created by the "参考总线(打包)" (RefBusOut) node - and fans it back out into the 48 fixed ports: 角色A–角色H, 场景A–场景H, 道具A–道具H, 音频A–音频H, 视频A–视频H, plus each video's soundtrack 视频A(音频)–视频H(音频).

In a workflow that stages a bunch of character/scene/prop/video/audio references, the pack side keeps the graph clean by collapsing everything to one wire; this node is where that wire becomes useful again, feeding individual reference images and audios into the encoding path.

How it works

Mechanically it's the inverse of the pack node: it takes the bus dict (the custom JZL_REF_BUS type, force-input so it only accepts a genuine bus), reads each of the 48 fixed keys, and returns them in the canonical order - characters, scenes, props, audios, videos, video soundtracks. Ports that were None at pack time come back None. It does no processing, no resizing, no scheduling; the docstring is blunt: "只做传输" - it only transports.

Outputs are the 48 fixed ports, typed by name: IMAGE for the character/scene/prop/video slots and AUDIO for the audio and soundtrack slots.

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 extra dependencies.

Common issues

The ordering matters. The 48 outputs follow the fixed BUS_KEYS order, so if you're building a workflow by port name rather than position, read the labels - 视频A(音频) (A's soundtrack) is not the same slot as 音频A (a standalone reference audio), and mixing them up is the classic way to get a silent segment with the wrong or missing sound. Also note this bus is deliberately incompatible with the slimmer JZL_REF2VA_BUS (18 ports) used by the ref2va encoder - the types don't cross-connect, which is the point. If your encoder only accepts ref_image_0…8 / ref_audio_0…2 style inputs, you want the pack's ref2va bus pair instead.

CategoryJZL/MiniMax

Inputs (1)

NameTypeDefaultDescription
busJZL_REF_BUS—

Outputs (48)

NameTypeDescription
角色AIMAGE—
角色BIMAGE—
角色CIMAGE—
角色DIMAGE—
角色EIMAGE—
角色FIMAGE—
角色GIMAGE—
角色HIMAGE—
场景AIMAGE—
场景BIMAGE—
场景CIMAGE—
场景DIMAGE—
场景EIMAGE—
场景FIMAGE—
场景GIMAGE—
场景HIMAGE—
道具AIMAGE—
道具BIMAGE—
道具CIMAGE—
道具DIMAGE—
道具EIMAGE—
道具FIMAGE—
道具GIMAGE—
道具HIMAGE—
音频AAUDIO—
音频BAUDIO—
音频CAUDIO—
音频DAUDIO—
音频EAUDIO—
音频FAUDIO—
音频GAUDIO—
音频HAUDIO—
视频AIMAGE—
视频BIMAGE—
视频CIMAGE—
视频DIMAGE—
视频EIMAGE—
视频FIMAGE—
视频GIMAGE—
视频HIMAGE—
视频A(音频)AUDIO—
视频B(音频)AUDIO—
视频C(音频)AUDIO—
视频D(音频)AUDIO—
视频E(音频)AUDIO—
视频F(音频)AUDIO—
视频G(音频)AUDIO—
视频H(音频)AUDIO—