Halftone Collage Motion Skill for AI Agents

Just ask

Say the word, it runs.

The skill fires on natural requests like these. No commands to memorize.

“Reverse-engineer this look and animate it”“Make a halftone paper-collage ad”“Break this reference down into a spec”“Turn this still into an assemble-from-empty motion graphic”

The pipeline

What the skill does for you.

1
Decode the reference
The skill over-analyzes a reference on purpose, naming the medium, exact color field, every cut-out element, its halftone and shadow treatment, the idea, and any burned-in label.
2
Emit an editable spec
It outputs a field-by-field JSON spec where every attribute is independently swappable, so 'same look, change the color field' rewrites one block and re-emits the whole thing.
3
Generate on-brand stills
Two stills per scene with nano-banana-2: an empty color field and the finished composition, labels burned in at generation (Advibly has no text-overlay tool by design).
4
Animate assemble-from-empty
Gemini Omni Flash opens on the empty field and the prompt drives the cut-out pieces sliding in and snapping into place with a crisp stop-motion cadence and native audio.
5
Lock the look and scale
Fix the color field, halftone, cut, and label convention once, then vary only the idea per scene so a set reads as one brand, not one-offs.
6
Assemble and publish
With a shell it concats the clips into one film; without one it hands back the clips in order, then can publish to a connected social account.

Paste a reference, get a spec back.

Every attribute of the look is a field you can change before a single image is generated.

Start creating

Decode, then vary

Any look, decoded into a spec.

Each set starts from a spec: the color field as a hex, the halftone density, the cut edge and its shadow, the label style. Lock it once and only the idea changes from scene to scene, so a set reads as one campaign. Change one field and the whole look moves with it.

Faithful by default: the decoded color field survives instead of being recolored to your brand
Labels drawn in at generation, never laid on top afterwards
9:16 or 16:9 on Gemini Omni Flash; 1:1 or 15 seconds moves the set to Seedance 2.0

Start creating Install the skill

Cobalt cut paper

Blue Tokai

Eye cream

The model stack

Right model, right step.

GPT Image 2 is the default for the stills: it follows a dense spec and burns the scene label in cleanly. If the dot pattern keeps flattening, the whole set moves to Nano Banana 2. Gemini Omni Flash is the default for motion, opening on the empty field and building the scene from there. When a clip has to land on an exact final frame, the set moves to a model that takes both ends, the empty still and the finished one. Your agent runs the newer models, like GPT Image 2.5 Sunburst, Gemini Omni Flash 1.1 and Seedance 2.5, when you ask.

What you get

A field-editable JSON spec of the decoded look
On-brand halftone stills with burned-in labels
Stop-motion, assemble-from-empty video clips
A cohesive multi-scene set that reads as one brand

Start creating Read the source

GPT Image 2.5 SunburstNew

The most detailed halftone stills when the set needs them

GPT Image 2.5 FlareNew

The fastest drafts of a scene while you settle the look

Seedream 5New

Dense layouts with native text, for a label that keeps breaking

GPT Image 2

Default: follows the field-by-field spec and burns the label in

Nano Banana 2

Keeps the dot pattern and cut edge when texture flattens, up to 4K

Gemini Omni Flash 1.1New

Builds from the empty field and can land on the finished frame

Seedance 2.5New

Exact empty-to-finished interpolation, 1:1, and clips up to 30 seconds

Gemini Omni Flash

Default: assemble-from-empty motion with native audio

Kling 3

Short teasers with both frames pinned and native audio

Install

Add it in seconds.

Install the skill, then make sure the Advibly MCP is connected so it has tools to call. Works with every skills-capable agent.

Claude Code

Claude

Cursor

Codex

VS Code

OpenClaw

Hermes

ChatGPT

Connect the Advibly MCP first

One click, no terminal

Download .skill Download .zip

Drop the .skill file into Claude → Settings → Skills, or unzip the folder into ~/.claude/skills.

One command (any agent)

npx skills add advibly/skills -s collage-motion

Works with Claude Code, Claude Desktop, Cursor, and any agent that supports skills. Pulls straight from GitHub.

Manual (Claude Code)

git clone https://github.com/advibly/skills.git
cp -R skills/collage-motion ~/.claude/skills/

Or in Claude Desktop: Settings → Skills → Install and select the skill folder.

Bundled references

The playbook.

Each skill ships with reference docs the agent loads on demand, so the craft is baked in.

collage-dimensions.md

Vocabulary for halftone, cut, shadow, color field, and label treatments, plus a fully worked decode example.

Honest limits

What to watch for.

Faithful by default (on_brand: false) so the decoded color field survives, not recolored to your brand unless you ask.
Every prompt is self-contained. Generators have no memory of earlier calls.
Labels are burned in at generation, never overlaid afterward.
Omni Flash has no end frame. For an exact final composition, 1:1, or 15s, the set switches to Seedance 2.0.

Questions

Before you install it.

What is a collage motion ad?

A short ad in the halftone paper-collage look: a flat bold color field, cut-out pieces printed in dots, clean edges with a soft shadow between layers and one sharp idea per scene. The motion builds the scene piece by piece, with the cut-outs sliding in and snapping into place, instead of wobbling a finished frame.

What do I have to give the agent?

A brand that is already onboarded in Advibly, plus a reference image to decode or a short brief. Store brands can pick a product from their catalog and its photo is used as the reference on every scene that shows it. If you mention an image but do not attach it, the skill asks for it rather than guessing.

How is it different from the Vox explainer skill?

Vox explainers are narrated stories built on a beat map, with a voiceover and a photoreal product shot. Collage motion is about the look itself: it decodes a reference into a spec you can edit, then makes a set of short scenes that assemble from an empty field, with native audio and no narration script.

How big is a set, and how long does it take?

Four scenes of 8 seconds by default, unless you ask for another count or length. You approve the stills before any motion renders. Each clip then renders in the background, and the set is joined into one video in the Advibly video editor, held on twos for a stop-motion feel, where you can fine-tune it.

What does it cost?

The skill itself is free and open source. Running it spends your existing Advibly credits on the stills and clips it generates, at no extra skill cost, and every generation shows its credit cost before it runs.

Which agents can run it?

Any agent that supports installable skills and the Model Context Protocol: Claude Code, Claude, Cursor, Codex, VS Code, OpenClaw, Hermes and ChatGPT. Install with one npx command, then connect the Advibly MCP so the skill has tools to call.