Zoey - 风格下拉选择 (@)

ZoeyStylePrompt is a prompt-magic dropdown for MiniMax H3's image-to-video mode. You pick one of 20 hand-drawn styles - sumi-e ink wash, cyberpunk neon, anime, ballpoint doodle, claymation, stained glass, gongbi, pop art, you get the idea - and the node hands you a complete, ready-to-paste prompt built around a very specific idea: a small animated character in that style springs to life inside your real photo and messes around with your subject. It's the "cute little drawing escapes its medium and plays with your dog/hand/product" effect, done as a structured, timed, one-take video prompt instead of you improvising a wall of text that H3 ignores halfway through.

It's part of comfyui-ZoeyTool's MiniMax H3 group, and its sibling ZoeyMiniMaxH3ReferenceToVideo is the node it's designed to feed. Where that node is the plumbing (models, reference slots, tag expansion), this one is the creative content.

How it works

One required input: style, a dropdown of 20 named styles. Output is the prompt - a long, fully-formed H3 I2V prompt - plus a style_label echo of the style you chose (handy if you're logging which style produced which clip).

What you're actually getting is a fixed narrative template, restyled per entry and timed out like a little screenplay. The structure is consistent across all 20 styles, which is the point of it: the uploaded reference image is the exact first frame and the visual anchor; your real subject never changes identity; only the small hand-drawn character appears, always with a fresh one-of-a-kind shape. Then it runs on a beat sheet - 0–3s they touch and react, 3–6s playful teasing and a leap onto the subject's shoulder, 6–10s it morphs while racing, 10–13s it escapes along a route, 13–15s it settles down near the subject for a warm ending. Each style swaps in its own visual language and SFX descriptions (ink splashes for sumi-e, 8-bit beeps for pixel art, etc.).

The optional inputs are the placeholders baked into that template:

• morph → [MORPH] (what it transforms into), escape_route → [ESCAPE_ROUTE] (how it gets away), final → [FINAL] (where/how it ends), mood → [MOOD] (the closing emotion). Leave any blank and the placeholder stays in the text - so you can fill it in later or deliberately keep it open.
• lock_background (default true) splices in an extra-strength instruction locking the first frame's scene, lighting, and camera so the background doesn't drift while the creature runs around.
• dual_ref - two-image version: image 1's real subject stays untouched while image 2 gets re-rendered into the chosen style and interacts with it. Off is the single-image "creature appears in your photo" mode.

Installing it

comfyui-ZoeyTool, standard:

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

Restart ComfyUI (or ComfyUI Manager → comfyui-ZoeyTool). Pure string output - no models here, but the node only earns its keep if you have MiniMax H3 installed to receive the prompt.

Where people get burned

First, this node outputs text - the "(@)" in the display name is a hint that it belongs to the @-reference MiniMax H3 workflow; you feed the prompt into your H3 image-to-video node's prompt field (or the pack's ZoeyMiniMaxH3ReferenceToVideo), you don't get video out of it directly. Second, the template is written for a 15-second, 16:9, one-take clip - if your H3 setup uses a different duration or aspect, the beat timings (0–3s, 3–6s…) no longer line up with the actual clip and the pacing gets weird. Third, the style prompts contain explicit negative-style instructions ("no 3D CGI, no subtitles, nothing on walls") that assume a photoreal base; if your "real subject" image is itself heavily stylized, the whole lock-background/subject-constant premise starts to fight itself. And remember these styles describe the character, not the whole frame - your scene stays photoreal, which is the joke the template is built around.