Just ask

Say the word, it runs.

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

“Make a Vox-style explainer for my product”“Paper / torn-paper collage animation”“Turn this claim into a motion collage”“Collage explainer ad”

The pipeline

What the skill does for you.

1
Write the beat map
The skill picks a narrative arc for the claim (problem-agitate-solve, how-it-works, myth-buster, and more) and drafts the full beat map: a hook that lands in under 3 seconds, two shots per beat, varied camera moves, and a narration line written to time. This is the one mandatory approval gate.
2
Pick the theme by eye
It proposes 3 to 4 theme presets that fit the claim's era and tone (american-retro, swiss-modern, punk-zine, paper-craft-cream, and more), renders beat one in each as a bake-off, and you pick the winner. That choice also settles the image model for the whole ad.
3
Generate the collage posters
One richly layered torn-paper poster per shot with nano-banana-2 for the paper grain, the headline baked in at generation (Advibly has no text-overlay tool by design), and your real product photo composited in photoreal, the one element never stylized.
4
Bring the collage alive
Each approved poster animates into living-collage motion with Gemini Omni Flash: layers parallax, edges flutter, pieces assemble and scatter, all SFX-only. The dramatic punches (assemble-from-empty, confetti, impact shake, whip) come from motion prompts alone. Seedance 2.0 handles exact end-frame reveals.
5
Narrate and score
The beats' narration lines join into one continuous read (xAI TTS) and an instrumental bed (MiniMax 2.6) sits under it, mixed over the paper foley with sidechain ducking so the voice always leads.
6
Stitch, caption, publish
The shots stitch into one Vox-style cut, optional dynamic captions can be burned in, and the finished explainer can post to a connected social account.

The model stack

Right model, right job.

nano-banana-2Torn-paper collage posters, 2K

gpt-image-2Type-heavy headline fallback

Gemini Omni FlashLiving-collage motion

Seedance 2.0Exact reveal / assemble-from-empty

xAI TTS + MiniMax 2.6Voiceover and score

What you get

The deliverable.

A finished Vox-style paper-collage explainer ad (9:16 or 16:9)
A collage poster per shot with the headline baked in
An approved beat map and a theme picked by eye
A narration script and score, mixed in
Everything filed in your Advibly library, ready to publish

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 command (any agent)

npx skills add advibly/skills -s vox-explainer

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/vox-explainer ~/.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.

beat-layer.md

The narrative arc library, hook patterns, beat counts, shot sizes, the flat-safe camera vocabulary, and the anti-monotony move rhythm.

prompt-guide.md

The 5-part image prompt structure, the dimension vocabulary banks, the 5-axis motion prompt, and the theme presets with the bake-off.

motion-collage.md

The dramatic looks (assemble-from-empty, hero fly-across, confetti, impact shake, whip) reached through motion prompts, no local scripts.

models-and-gotchas.md

Image and video model choice, content blocks, the positive-phrasing rule, and the full final-mix ffmpeg recipe.

Honest limits

What to watch for.

The look lives in the poster. If a poster is not a rich layered collage, motion cannot save it, so weak posters get re-rolled before animating.
Every headline is baked in at generation; there is no text-overlay step. A degraded headline is re-rolled or shortened.
The real product stays photoreal while everything else is paper. Redrawing the product as paper is the most common failure and gets caught before motion.
One image model and one video model per delivered ad; the set swaps whole or not at all.
Real people or third-party logos move the whole ad to Kling, since the default motion models block recognizable faces and marks.

More skills