About this template
Ship a Vox-style ad without writing prompts
Vox-style paper-collage explainers are pulling millions of views on TikTok and Instagram. Media buyers are running them as paid creative. They work because they read as editorial — not as another UGC talking head.
Making one used to mean a motion designer or days in After Effects.
This agent is the whole pipeline. You talk about your product for about 30 seconds. It writes the voice-over and visual direction, waits for one approval, then either generates the clips for you or hands you finished Omni Flash prompts.
No prompt copy-pasting. No "extract the last frame yourself." No babysitting a chain of tools.
Who this is for
You're a fit if:
- You need a 30-second "here's how this works" ad (app, supplement, SaaS, gadget)
- You want the editorial collage look (torn paper, labels, arrows) — not selfie UGC
- You'd rather brief an agent once than run Seedance / CapCut gymnastics by hand
You're not a fit if you want a photoreal talking-head UGC actor. That's a different template.
What you get the moment you fork
vox-style-adskill — intake → script + visuals → generate- Mode A (autopilot) — MaxFusion AI MCP + Google Omni Flash, last-frame chained clips
- Mode B (manual) — same creative system, finished prompts you paste into your own tool
- Style bible, 30s beat map, Omni prompt templates, CapCut → Replace Voice handoff
Why Omni Flash (not Seedance)
A Vox explainer lives or dies on words on screen and words spoken.
Seedance is slower, more expensive, and weak at both. You regenerate. You burn credits. You throw clips away.
Omni Flash holds text, holds speech, holds style across clips, and finishes fast. For this format it isn't close.
How a job runs
- Tell it the product (image optional), mascot vs faceless, vertical vs horizontal.
- It shows one approval block: full VO script + per-clip visual direction.
- You say go.
- Mode A generates ~10s clips and chains the last frame into the next. Mode B gives you the paste-ready prompts + chain checklist.
- Stitch in CapCut → MaxFusion Replace Voice once for a consistent narrator.
You come back to a finished ad — or a package you can finish in one sitting.
Say this after you fork
Make a Vox-style ad for my product: <name> — <one-liner>.
Faceless, vertical, 30 seconds. CTA: <action>.
Or: run the example demo for a fictional sample product.
Bring your own keys (nothing is baked in)
- Claude Code — bring your own model credentials on the forked VM (none ship in the template).
- MaxFusion (optional, Mode A) — your own MaxFusion account. Open MCP settings, use Copy instructions for LLM, and wire MaxFusion MCP into Claude Code from your side.
- No MaxFusion? Use Mode B. Same creative system.
Product images and mascot art are per-job uploads. No credentials ship in the template.
The difference vs "prompt pack" skills
Other skills hand you prompts and tell you to generate, extract the last frame, generate again, then edit outside.
This one is built for MaxFusion MCP + Claude Code: you chat for 30 seconds and the agent runs the chain. Or it drops finished prompts if you stay manual.
Fork it. Brief your product. Approve once. Ship the explainer.