AICG3D 提示词模板

AICG3D_PromptPreset - AICG3D 提示词模板 - is the least mysterious node in the pack and, judging by how often people ask what's in a 144-item dropdown, the most confusing to look at. It reads from the pack's bundled prompt_presets/ folder and outputs the text of the template you picked. That's it. It's a string node with a library attached.

Why you'd want one: H3 prompts for film work aren't one-liners. The good ones in this pack's presets run to a page of structure - shot framing, lighting, timing, dialogue pacing, what to keep static. Typing that into a widget every time is how you end up with inconsistent output across a ten-segment chain. Pick the template once, chain the node, and every segment inherits the same directorial rules.

What's actually in the library

The dropdown is 分类｜名称 - think H3 专题｜H3 单图角色 PV 提示词模板（盲盒万能版）. The categories cover a mix of things: H3-specific single-image character promos, image-to-video multi-shot systems, official H3 case prompts (fashion ad, retro Japanese anime title sequence, Chinese kinetic typography, cyberpunk character UI, coffee-into-desert seamless transition, magician costume swap, dusk train hand-drawn lighting), a fight-choreography skill, and general 3D-animation assistant prefixes. Plus the usual style/storyboard/portrait/eastern/image/video groupings.

It's a genuinely odd grab bag, and that's the honest description: it's a curated dump from the author's own working collection, not a designed taxonomy. Type a keyword in the widget to filter - H3, 打斗, 广告, 分镜 all narrow it fast.

How it composes

Pick a preset and leave task empty and you get the template body with a header line naming it. Fill task in, or drive it from upstream through the optional task_override input, and you get:

===== 提示词模板：<name>（<category>）=====
<the template body>

===== 任务 =====
<your task>

Three outputs: 提示词 (the composed string - this is the one you wire), 模板ID (the preset's stable id, useful for logging which template a render used), and 模板正文 (just the body, no header, if you want to wrap it yourself).

Wire 提示词 into the prompt input of a segment node, or into the main H3 node, or into a dialogue/LLM node if you're running the optimiser chain.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/JGRFW/comfyui-AICG3D

Restart, hard-refresh, category AICG3D - or via Manager, searching AICG3D / 短剧工作台 / MiniMax H3.

There's no model, no download, no dependency. It's reading markdown off disk.

Things worth knowing

The dropdown is built at startup. If you add your own template to prompt_presets/ (or edit one), restart ComfyUI before you expect to see it. Editing a file in place without a restart gives you the old text, which looks like the node ignoring you.

Don't use it as a prompt-enhancer. It doesn't call a model and it doesn't rewrite your idea - it concatenates. If you want the LLM actually thinking about your prompt, that's the prompt-optimiser settings panel, and if you want the model to pick a creative framework for you, that's AICG3D 技能加载器. This node is the "I already know what structure I want" option.

Templates assume H3. They're written with H3's behaviour in mind - native 24fps, clip lengths, audio baked into the generation. Feeding a 200-word H3 cinematic-grammar prompt into a different video model will not produce a better clip; it'll produce a confused one. Pick the preset to match the model, not the other way round.

Long bodies + the optimiser can fight. If you run a preset through the prompt optimiser, the optimiser may compress or restructure the template's careful sequencing. Often that's fine. When a shot gets noticeably worse after optimisation, take the optimiser out of the chain and try the raw preset before you go rewriting the template.