MiniMax H3提示词扩写指令 V2·简单基础版（Tlant）

If the V2 Full Prompt node is the kitchen-sink version of the Tlant MiniMax H3 prompt family, MiniMax H3提示词扩写指令 V2·简单基础版（Tlant） is the pared-down countertop model: just the settings most people actually change, nothing hidden, nothing random, and - crucially - nothing silently added.

Here's the one idea to internalize: in V2, what you don't see is what you don't get. The simple version doesn't secretly fill in the Full Prompt's options behind your back. An element that isn't displayed here isn't written into the instruction, ever. If you want depth-of-field control or beat-synced cuts, you need the Full Prompt node. If you want to set a handful of things and ship, this is the one.

It uses the same three-valued system as its big sibling, applied to every dropdown:

• 不指定 (default) - the element stays out of the instruction entirely.
• 无 - explicitly banned; for unavoidable properties, "don't deliberately change the source."
• 自行推断 - let the remote LLM judge from the image and theme.

No local seed, no random fill. Repeat your settings and you repeat your output.

What's exposed here is a sensible "common essentials" list: 面部朝向 and 视线行为 (face direction and gaze), 表情 (expression), 切镜模式 (cuts), 结尾方式 (ending), 运镜类型 (camera move, mapped to H3's official vocabulary), 背景锁定 (background), 头发运动 and 服装运动 (hair and fabric), and 环境事件 (an environmental event for background life). The audio block covers 音频模式, 环境声预设, 对话预设, 声音风格预设, plus the three custom text boxes - 自定义环境声, 自定义对话, 自定义声音风格 - each of which completely overrides its preset when non-empty. Then 声音语言 and 口型同步 round it out.

The two protection policies carry over from the Full Prompt node and are worth repeating: music is opt-in (only an explicit music selection, a concrete music parameter, or "infer" lets the LLM add background music; otherwise non_diegetic_music is forced to N/A), and dialogue is off by default - no invented speech, singing, narration, or on-screen text unless you ask for it. That's the whole "no surprises" promise in action.

Common inputs with the rest of V2: 视频时长 (4–15s), 主题预设/自定义主题 (custom theme overrides the preset outright), and 提示词长度最小值 (default 200 - the minimum English words the final H3 prompt must contain, reached through detail, never padding).

Outputs: LLM扩写指令 (STRING) and 图像 (IMAGE passthrough). Wire both into a vision LLM - Tlant's own Llama Server Chat is the intended bridge - and that LLM returns the clean, final MiniMax H3 I2VA prompt for your video workflow.

Install

ComfyUI Manager → search ComfyUI-Tlant-Toolkit → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Tlant/ComfyUI-Tlant-Toolkit

Restart ComfyUI. No dependencies, no model downloads.

Gotchas

Same two as every H3 node here: you must connect an image or supply a non-empty 图像描述, or the node refuses to run. And if a setting you expected is missing from the canvas, that's the design working - it's not in the simple node, so it's not in the prompt. Reach for the Full Prompt node rather than assuming this one handles it invisibly.