JZL - 🔗 MiniMax H3 ref2va参考总线（打包）

"JZL - 🔗 MiniMax H3 ref2va参考总线（打包）" (ref2va Reference Bus Pack) bundles the 18 fixed reference ports that the pack's "MiniMax H3 参考编码2" node (and the official H3 ReferenceToVideo node) expects, into one wire. Nine reference images (ref_image_0–ref_image_8), three reference videos (ref_video_0–ref_video_2), the three matching video soundtracks (ref_video_audio_0–ref_video_audio_2), and three standalone reference audios (ref_audio_0–ref_audio_2).

If you're routing a handful of reference assets into ref2va encoding, this is the tidier way to do it - one output instead of up to eighteen dangling wires, and a custom type that stops you plugging it into the wrong node.

How it works

A pure dict packer. Every input is optional; the port types are inferred from the names (audio-bearing ports take AUDIO, the rest IMAGE), and pack gathers whatever is connected into a dict keyed exactly as the ref2va encoder's inputs are named. That naming is the entire point of the design: the source is explicit that the interface names are "完全对齐" - fully aligned - with the encoder's input names, so on the other end you can unpack and connect straight to the encoder without renaming.

The output is a single ref2va参考总线 of the custom JZL_REF2VA_BUS type. Note it's deliberately a different type from the pack's 48-slot JZL_REF_BUS - the two won't cross-connect, which prevents the "wrong bus into the wrong decoder" failure class entirely.

How it fits

This is the manual, wire-based alternative to the pack's wireless asset pool. If you're hand-building a workflow with specific reference images/videos/audio, route them here, then use the sibling unpack node before the encoder. Pair it with "参考编码2" (the fixed-port ref2va encoder) - the pack's dynamic Autogrow version takes its own growing inputs and doesn't need the bus at all.

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; transport only.

Common issues

Unconnected inputs become None in the bus and come back None on the other side - a missing reference is silent, so check the pack end when a segment ignores a reference. And keep the count limits straight: at most 9 images, 3 videos, and 3+3 audio tracks fit in this bus, which mirrors what the ref2va encoder itself accepts. The videos should be frame sequences (IMAGE) and their soundtracks must land on the matching ref_video_audio_N port - that pairing is positional, and mixing it up gives you a clip with the wrong audio.