JZL - 🎧 音频调度2

"JZL - 🎧 音频调度2" is the fixed-port sibling of the pack's dynamic "音频调度". Same job - read the Script Processor's audio scheduling and route the right reference audio to each H3 segment - but instead of letting you hang however many dynamic AUDIO inputs you like, it hands you eight fixed, named inputs: 音频A through 音频H. If you value a stable, wireable layout over flexibility, this is the one you want.

How it works

The required va_instruction (*, force-input) carries the audio scheduling string from the script/segmented-text chain. The node parses it into an ordered slot list, and for each slot matches by fixed interface name - write 音频A in the instruction and it picks the value off the 音频A input. Matched audios come out in slot order on ref_audio_0, ref_audio_1, ref_audio_2, wired to the ref2va encode path.

Two routing sources, in priority order:

1. The global asset pool ("无线传输" / wireless transfer) - the pack's Generation Manager or asset manager writes named assets into a shared pool, and the dispatcher pulls by name first, with no wire at all.
2. The fixed 音频A–音频H inputs - fallback for older workflows wired the classic way.

That dual path is the same pattern the pack's scene and video dispatchers use: it decouples "which assets exist" from "which segment uses which", which is the whole point of the pool. Output order follows the instruction's slots, and like the dynamic version, empty instructions emit None rather than guessing.

The inputs that matter

• va_instruction - the audio scheduling string.
• 音频A … 音频H - eight fixed AUDIO inputs. Only three can be emitted per run (ref_audio_0/1/2), matching H3's ref2va limit.

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; it's pure routing logic.

Common issues

The fixed-name matching means exactness matters: the slot has to say 音频A, not 音频1 or A. If a segment goes out with no audio, verify the instruction literally contains the interface name. Also note the "2" in the name marks the fixed-port version, not "two audios" - the limit of three ref audios still applies because that's what the downstream ref2va encoder accepts. And the usual pool gotcha: if you rely on the wireless pool, the asset manager must have run in the same session, because the pool is in-memory and rebuilt each queue.