JZL - 💾 视频保存分配

When you generate a multi-segment 漫剧 (comic drama) with the JZL MiniMax pack, the 🗂 MiniMax-H3生成管理器 (generation manager) runs all your H3 prompts and hands you one big result: a generation bus containing every segment's frames and audio. But you don't want one blob. You want each segment as its own (video, audio) pair so you can preview it, keep it, retry it, and finally stitch the whole thing together. JZL - 💾 视频保存分配 (Video Save Distributor) is the node that unpacks the bus into those individual pairs.

How it works

It's the output side of a "wireless bus" pattern. The generation manager writes each segment's IMAGE + AUDIO into an in-process asset pool (JZL_BUS_POOL), and this node reads them back by group number - no giant wire dragging tensors across your canvas. You connect the bus, it counts the groups, and it fills as many of its 12 output pairs as the bus actually produced:

• 图像1 (IMAGE) + 音频1 (AUDIO) … up to 图像12 + 音频12

Each pair feeds a Video Combine (or any save/preview node), so segment 3 goes to its own combine while you're still deciding whether to keep segment 2. That per-segment isolation is the entire point - retry one shot without regenerating the whole episode.

Inputs and the one gotcha

• bus (STRING) - connect the generation manager's 生成总线 output. That's the only input; the tooltip spells it out.

The gotcha is the placeholders. For groups the bus didn't produce, the node outputs a harmless 1×32×32 black image and a silent audio stub rather than None - explicitly so downstream nodes don't crash. That's convenient, but it means an empty output port doesn't look empty. If a segment's combine is spitting out a black frame, check whether the bus actually contained that group, or you'll spend a while debugging a "silent failure" that's really just a segment that failed to generate upstream.

What it's for

Two jobs, really. The obvious one is finishing: distribute segments → Video Combine → save, and your episode is one file. The less obvious one is selective finishing - you can wire combines only for the segments you're happy with, or route different segments to different output formats. In a story where the LLM wrote 16 segments and three of them came out wrong, this node is what lets you regenerate those three in isolation instead of re-rolling the dice on all 16.

Install

Standard JZL pack install: ComfyUI Manager → search JZL or ComfyUI-JZL-MiniMax-H3, or git clone https://github.com/wjluoxiao/ComfyUI-JZL-MiniMax-H3 into custom_nodes/ and restart. Needs a recent ComfyUI with the official H3 nodes. It's pure routing - no models, no LLM runtime. If you're combining video with audio, make sure your Video Combine chain handles the AUDIO input this node provides; and remember this pack is Chinese-first, so the output ports are labeled 图像 (image) and 音频 (audio).