⚡ Aimacor Labs — Video Prompt Director Lite (EN)

If you've ever hand-written a Wan 2.2 image-to-video prompt, you know the pain. Video models want flowing English prose about motion - "she slowly turns her head and lifts one shoulder while the camera dollies in" - not the comma-tag soup that works fine for SDXL. Type it by hand and you'll fight yourself: "tilts" one clip, "tilted" the next, you forget the camera move, and the lighting block you swear you wrote keeps quietly disappearing. That's the gap this node fills.

Video Prompt Director Lite (EN) is a prompt-builder, not a model. You pick a head position, an expression, a hand pose, a camera move from dropdowns, and it assembles a coherent English prompt for you. It's the English-UI half of a bilingual pack from Aimacor Labs (GitHub: titomatus0203/aimacor-labs-video-prompt-director-lite). Fair warning up front: this is a brand-new, very small pack - the repo had a single commit and zero community track record when I checked. It's a useful utility, not a battle-tested one. The name is also a lie in the good way: it calls no API and needs no key.

How it builds the prompt

Under the hood it's dead simple, and that's the appeal. The node keeps a set of editable JSON "libraries" - libraries/head.json, hands.json, camera.json, and so on - where each entry is a display name plus a ready-made English phrase, like "She slowly turns her head to the left in one smooth, continuous motion."

When you queue it, it stitches those pieces together: an intro line (Ultra-realistic cinematic video of {character} {scene_context}), then one phrase for each body part you set, then a temporal-continuity block the code literally labels "WAN 2.2 ANTI-LOOPING", then your camera move, then a lighting-and-quality block ("realistic lighting consistent with the reference image... fashion editorial quality, 4K resolution"). The reference-image wording gives away its target: this is tuned for Wan 2.2 image-to-video, which is exactly the community's default character-video workflow. Every phrase in the libraries is written as "She...", and if you set Gender to Male the node regex-swaps her/she/her to him/he/his. The output is always English regardless of which UI variant you use - Wan's encoder wants English prose.

The inputs that actually matter

Fifteen inputs sounds intimidating until you realize almost all of them are dropdowns that default to NONE. The ones to set:

• Character (LoRA Trigger) - the name or trigger word for your character LoRA. This is the spine of the whole thing: it's what keeps your person consistent across the clip.
• Scene Context - free text describing the setting ("standing in a sunlit studio"). Nice touch: if you type "She is standing...", it strips the "she is" prefix so you don't duplicate the character in the intro.
• Expression, Gaze, Hands, Camera - the four dropdowns that do the heavy lifting for performance and framing.
• Manual Prompt (optional) - the escape hatch. Per the author's own placeholder: when this field has text, all the automatic controls are ignored and it returns your text untouched.

The single output, prompt (STRING), wires into a CLIP Text Encode (Positive) feeding your Wan 2.2 I2V pipeline.

Installing it

No dependencies, no model downloads, nothing. requirements.txt is empty and the Python is pure stdlib. Either search ComfyUI Manager → Install Custom Nodes for aimacor-labs-video-prompt-director-lite, or:

cd ComfyUI/custom_nodes
git clone https://github.com/titomatus0203/aimacor-labs-video-prompt-director-lite

Restart ComfyUI and the node appears under Aimacor Labs/Video Prompt Director.

Gotchas

• Manual Prompt silently wins. If you leave text in that box from an earlier experiment, your dropdowns do nothing and you'll think the node is broken. Clear it.
• English output on a Spanish-UI pack - the sibling node Video_Prompt_Director_ES translates the labels, never the generated prompt. That's intentional.
• It's opinionated about tone. "Ultra-realistic cinematic" and "fashion editorial quality" are baked in; if your style is stylized or anime, the intro and quality blocks may fight you. Since the libraries are plain JSON, you can edit the phrases yourself - which is honestly the most interesting part of this pack.