JZL - 📋 列表分发

"JZL - 📋 列表分发" is the workhorse that turns one giant script into many usable lines. In the JZL-MiniMax-H3 pack's pipeline, the "剧本与镜头处理器" (Script Processor) LLM writes a whole story as a block of [SHOT_START]…[SHOT_END] segments, and this node carves that block into per-segment chunks, pulls out one field from each, and pushes them to however many output ports you want. It's the node that makes "one story, six separate video generations" actually tractable.

How it works

The input is a 文本输入 (STRING, force-input) that comes straight off the script processor. The node splits it on [SHOT_START]…[SHOT_END] blocks, then inside each block looks for a field marker and extracts that field's full content - without breaking lines, per the source. Which field it pulls is decided by 调度类型:

• H3提示词 - the H3_PROMPT field (default)
• 场景调度 - SCENE_INSTRUCTION
• 视频调度 - VIDEO_INSTRUCTION
• 音频调度 - AUDIO_INSTRUCTION

Then it fans the results out to the first 输出数量 of its up-to-99 STRING outputs, named 分段1…分段99. One quirk worth knowing: if the input has no [SHOT_START] tags at all - say a downstream node already extracted the list - the node treats each incoming list element as one segment and passes it through, so it doubles as a generic list fan-out. It's declared INPUT_IS_LIST, so it accepts both a scalar string and a list.

The inputs that matter

• 调度类型 - which field to extract per segment. This is the one you'll actually change.
• 输出数量 - number of populated outputs, 1–99, default 4.

Everything else is plumbing. The outputs are plain STRINGs, so anything downstream that eats text will take them.

How it fits

You'll typically see it between the Script Processor (or the "分段处理中心" ShotFormatter) and the encoding/scheduling layer: extract H3 prompts per segment and feed each into a MiniMax H3 参考编码 or the asset manager, or extract the 场景调度/音频调度 fields to drive the scene and audio dispatchers. It's a pure text-formatting node - no model, no GPU, no weights.

How to install it

Part of the same pack as everything else - ComfyUI Manager, search "ComfyUI-JZL-MiniMax-H3", or:

cd ComfyUI/custom_nodes
git clone https://github.com/wjluoxiao/ComfyUI-JZL-MiniMax-H3

Restart ComfyUI. This node is pure Python stdlib (re), so it needs no extra dependencies at all - a nice contrast with the rest of the pack.

Common issues

The most common mistake is forgetting that the segment count you request has to match the actual number of [SHOT_START] blocks; leftover outputs just come out empty, and an empty STRING flowing into a video encoder gives you a confusing "no prompt" failure a few nodes later. Also, if your upstream script doesn't use the exact [SHOT_START]/[SHOT_END] markers the node scans for, you'll get the fallback passthrough instead of split segments - check the field markers if nothing comes out the way you expected.