JZL - 📝 MiniMax H3 手写提示词

"JZL - 📝 MiniMax H3 手写提示词" (Handwritten Prompt Editor) is the pack's no-LLM escape hatch: a plain multiline text box where you type an H3 prompt yourself, with one genuinely useful trick - a "➕ 添加元素" button that drops in H3's six-segment field, scheduling-command, and storyboard templates at your cursor. No Script Processor, no model loader, no API key. Just you and the prompt grammar.

It exists because the pack's normal path is LLM-generated scripts, and that's not always what you want. Sometimes you have a one-off shot, a specific idea, or a paid prompt you don't want a model to "improve." This node is the direct line into the encoding pipeline for those cases.

How it works

Mechanically it's nearly trivial - a prompt_text STRING in, the same string out. The real work is in the front-end: the insert button runs js/h3_prompt_editor.js, which understands the H3 six-segment prompt grammar and inserts correctly-formatted blocks (field markers, [SHOT_START]/[SHOT_END] segmentation, scheduling directives) at the cursor instead of making you remember the exact syntax. You write; it formats.

Output is a single H3提示词 STRING. Per the source docstring it's designed to feed the "分段处理中心" (ShotFormatter) shot_text input or the "提示词增强" (Prompt Enhancer) node's script_text input - or it can go straight to an encoder node if you're bypassing the chain.

The input that matters

• prompt_text - the multiline box. The tooltip says it all: handwritten H3 prompt; click "➕ 添加元素" to insert templates at the cursor.

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 ComfyUI so the front-end JS loads. No Python dependencies beyond the pack.

Common issues

Two things worth knowing. First, the node is a pass-through: it won't validate your prompt or warn you when the grammar is wrong - a typo'd marker just rides through and H3 does its best. Second, the templates insert assuming you're writing per-segment H3 prompts; if you paste a whole multi-shot story in here you're expected to format the [SHOT_START] blocks yourself, because downstream nodes split on those markers. Also keep the model context in mind: H3's weights are ~42.5GB and the MiniMax H3 Community License excludes the US, EU, UK, and South Korea from local runs - this node, at least, costs you nothing.