Zoey - Wan2.2提示词生成器

WanPromptGenerator is a structured prompt builder aimed squarely at Wan 2.2 video generation - the model that is still, as of now, the quality king of local text-to-video. If you've tried prompting Wan, you know the drill: it rewards long, cinematic descriptions - subject, scene, motion, camera language, lighting, shot size - and punishes short tag soup. Writing those by hand for every clip is where ideas go to die. This node hands you sixteen structured controls (thirteen dropdowns and three free-text fields) and assembles the comma-joined, film-script-style prompt for you. It's from comfyui-ZoeyTool, a pack clearly built by someone who generates a lot of Wan video and got tired of typing the same cinematography boilerplate.

Here's the honest framing: it won't write a better prompt than you could. It makes you build one faster and more consistently, and it keeps your experiments reproducible because every knob is a visible setting. That consistency is the real value - prompt-building nodes like this exist precisely because model quality is gated on prompt structure, and structured pickers remove the variance.

How it works

Thirteen dropdowns plus three free-text fields - sixteen knobs total. The three free-text fields carry the substance: 主体 (subject), 场景 (scene), and 动作 (action) - each with an optional 主体细节 / 场景细节 / 动作细节 detail box that gets appended after a comma (defaults describe a girl in a field with flowers, which is the author's demo scene, so change them). Everything above them is a dropdown.

The dropdowns map to real cinematic vocabulary Wan responds to:

• Light: 光源类型 (daylight/artificial/moonlight/practical/fire/fluorescent/overcast/mixed/clear-sunlight) and 光线类型 (soft/hard/top/side/back/bottom/rim/silhouette/low-contrast/high-contrast).
• Camera: 景别 (shot size: close-up→wide), 构图方式 (composition), 镜头焦段 (focal length), 机位角度 (angle), 镜头运动 (camera motion: push/pull/pan/手持/环绕…), 镜头类型 (single/two/three/group/establishing shot).
• Look: 时间段, 色调风格, 画面风格 (felt/3D-cartoon/pixel/puppet/clay/anime/watercolor…), 面部表情, and 主体动作.

Every dropdown includes 无 as an option, and "无" options are simply dropped from the final string - that's the mechanism that lets you use only the controls that matter for a given shot. The output, 提示词, is one comma-joined string ready to wire into Wan's text encode (or the Wan2.2 prompt slot in whichever wrapper you run - Kijai's wrapper and native ComfyUI both eat this format fine).

Installing it

Standard pack install - comfyui-ZoeyTool:

cd ComfyUI/custom_nodes
git clone https://github.com/liangzoey/comfyui-ZoeyTool.git
cd comfyui-ZoeyTool
pip install -r requirements.txt

Restart, or ComfyUI Manager → comfyui-ZoeyTool. This node is pure string assembly - no models, no downloads.

Where people get burned

The trap here is scope: this builds a shot prompt, not a negative prompt, and it has no notion of Wan 2.2's two-pass architecture or the speed-LoRA tradeoffs people fight about (keep that debate out of the generator; the community consensus is speed LoRAs hurt motion and emotion, but that's a sampler/model question, not a prompt one). Second, the UI is entirely Chinese - if you don't read it, remember 无 = none/omit. Third, its defaults are a fixed demo scene; first thing you do is replace 主体/场景/动作, or every clip you make is a girl in a field. And a friendly reminder that Wan 2.2 rewards detail density: use the three detail boxes - the dropdowns give structure, but the free text is where the specificity that separates "usable" from "great" actually comes from.