meta-image-ad-generator: Meta image ads

Ad creative Desktop to render

What it does

meta-image-ad-generator creates Meta image ads, from concept through finished image file. You give it a brand and an offer, it researches your buyer, extracts your brand style, and produces a batch of image-generation prompts tuned to specific personas, awareness levels, and psychological triggers. On the desktop app, it goes one step further and renders the finished image files for you.

It handles the decisions that eat time: which persona to target, which cognitive bias to lean into, which visual style fits the angle, and how to phrase the prompt so the image model produces something on-brand rather than generic. You end up with a set of ad concepts ready to test, backed by a research pass rather than gut feel.

When to reach for it

You need fresh Meta image-ad creative for a brand and don’t want to start from a blank prompt.
You’re launching a new campaign and want concepts grounded in real buyer psychology.
Your current ads have gone stale and you want new angles to test against your control.
You want to work through a persona systematically, testing one buyer type at a time.
You have a specific concept in mind and want a polished, on-brand prompt for it rather than writing one from scratch.

Where it runs

Cowork: concepts only Desktop: full render

The concepting and prompt-writing steps work in both Cowork and the desktop app. You can research your buyer, build personas, select angles, and produce polished image-generation prompts from either environment.

Rendering the finished image files calls an outside image service (Nano Banana / Creatify). Cowork’s sandbox blocks that API call today, so to get actual image files out you need the Claude Code desktop app with your image API keys configured. See API key setup and choosing your setup for details.

How to run it

Describe what you want and name the brand:

“Create Meta image ads for examplebrand.com, targeting people who are deal-comparing and almost ready to buy.”

Or call it directly:

“/um-toolkit:meta-image-ad-generator for examplebrand.com, cold traffic, lifestyle angles”

Once running, you can take one of three paths:

Guided selection: walks you through persona, awareness level, cognitive bias, and visual style step by step.
Custom concepts: you describe specific ad ideas, it maps them to the right triggers and styles, then generates prompts.
Quick generate: auto-selects the best persona and settings and returns five prompts immediately.

What it needs from you

Your brand URL (or an onboarded brand folder, if you’ve already run setup).
A product reference image for any ad that should show your actual product. Without a real reference, the image model produces a lookalike with the wrong label and colors. One good product photo is all it needs.
Your offer or angle, even roughly: “we’re promoting the starter kit at $49” or “we’re testing a social-proof angle.”
Image API keys configured in the desktop app if you want rendered image files, not just prompts.

What you get back

From either environment:

A set of ad concepts, each tied to a specific persona, awareness level, and cognitive bias so you know why the angle was chosen.
Image-generation prompts, ready to paste into Nano Banana or run through Creatify: complete with brand-style injection, safe-zone placement specs, and text overlay instructions.
A campaign state file saved to Google Drive so you can resume the same brand later without starting over.

On desktop (with API keys):

Rendered image files for each prompt, produced through Creatify’s batch pipeline.

A worked example

You: “Create Meta image ads for examplebrand.com. We sell a magnesium supplement for sleep. Target people who’ve tried other sleep supplements and are skeptical.”

meta-image-ad-generator: (after researching the brand and building personas)

Persona selected: Skeptical Self-Optimizer, 38-52, has tried melatonin and gummies, stopped because of grogginess. Knows the category, doesn’t trust bold claims.

Awareness level: Solution-aware. They know magnesium exists for sleep; they don’t know why yours is different.

Angle chosen: Social proof with specificity bias. Real testimonials with precise language outperform generic “sleep better” headlines with this persona.

Prompt 1 (lifestyle social proof): “Render in 9:16 portrait (1080x1920). Centered 4:5 safe zone. Warm bedroom, phone-camera quality. Woman in her 40s, natural imperfections, reading on a nightstand. Text overlay inside safe zone: ‘I’ve tried everything. This is the first thing that actually worked.’ Brand palette applied: [brand guide injection]. All critical content within 4:5 safe zone. Text legible on phone at arm’s length.”

(Four more prompts follow, each on a different visual style and secondary angle.)

Note: the above is an illustrative example of the style and structure you’d receive. Your actual output will reflect your specific brand, product, and research.

Tips and gotchas

A real product photo is the single biggest quality lever. When the skill has a reference image, the image model renders your actual product with the right label, color, and packaging. Without it, you get a convincing-but-wrong lookalike. Drop your best product photo into the brand folder before you start.
Rendering needs the desktop app and API keys. If you’re in Cowork, you’ll get polished prompts you can use manually. Finished image files require the desktop setup. See the setup guide.
The safe-zone rules are applied automatically. Every prompt this skill generates follows the 9:16 canvas and 4:5 safe-zone standard so your ad runs cleanly across Meta Feed, Reels, and Stories without cropping. You don’t need to add placement specs yourself.
Campaigns save and resume. The first time you run it for a brand, it builds full research and personas. After that, it loads from your saved campaign state so you can pick up where you left off without repeating the initialization.

apollo: the market research skill that meta-image-ad-generator runs behind the scenes during initialization. Run it separately if you want the full research brief first.
ad-overlay-copy-engine: generates the exact text overlay copy for your ads once you have your prompts.
creatify-batch and image-sourcer: handle the actual rendering pipeline on the desktop. meta-image-ad-generator hands off to these automatically.
seedance-video-ad-generator: the video equivalent, for when you want the same research-backed, persona-driven approach applied to short-form video ads.