How to Steal the World’s Best Ad Creative (Legally) Using AI

At an AI workshop in January, I pulled up a Louis Vuitton ad on the screen.

I pasted it into Gemini and typed one instruction: “Reverse-engineer this photo into a JSON prompt that I could use to recreate this image with a different product.”

Gemini gave me the full breakdown. Composition style. Lighting setup. Lens approximation. Color grading. Mood reference. The framing choices that made the shot feel expensive.

Then I changed one word in the prompt. Swapped “biker jackets” for a different product. Ran it through an image model.

Same aesthetic. Different product.

The room went quiet for a second. Then someone said, “So the best ads in the world are basically a free prompt library.”

Yeah. Exactly.

Why Great Ads Look Great

Most people assume expensive-looking product photography comes down to budget. A great photographer, a good set, professional lighting equipment.

Budget helps. But what actually makes a great shot is a set of composition decisions: the kind of shot, the kind of light, the angle, the negative space, the mood. These decisions are learnable. And more importantly, they’re extractable.

When you look at a Louis Vuitton ad, you’re looking at decades of accumulated visual intelligence. Every element of that photo was a deliberate choice made by someone who spent years studying what works.

AI can read that photo and tell you what those choices were.

Camera Language: What Most People Miss

There’s a concept I teach called camera language. It comes from the idea that when you’re prompting for images, the language of photography (not the language of marketing) is what drives quality.

Generic prompt: “A woman wearing a jacket, professional look, nice lighting.”

Camera language prompt: “85mm portrait lens, soft directional key light from upper-left, slight underexposure for richness, shallow depth of field, editorial fashion framing, cool-muted color grade.”

The second prompt tells the AI what a camera operator would know. Focal length, light source, exposure choice, depth. The output looks completely different: more intentional, more commercial, less like generic AI output.

Most people don’t know camera language. And that’s fine, because you don’t have to learn it yourself. You can ask the AI to extract it from a photo you already love.

Reverse Prompting: Working Backwards

I started teaching this as “reverse prompting.” Instead of trying to describe what you want from scratch, you start with a result you already like and work backwards.

The process:

1. Find an ad or product photo that has the look you want.
2. Paste it into Gemini (or any vision-capable AI).
3. Ask: “Reverse-engineer this image into a detailed prompt I could use to recreate this style with a different subject.”
4. Take the JSON or text prompt it gives you.
5. Change one element (your product, your model, your setting, your season).
6. Run it through an image model.

What you end up with is a style-matched image for your product that would have cost a real photoshoot to produce.

Building a Prompt Library

The real power isn’t the one-off technique. It’s what happens when you start doing this systematically.

Say you’re selling luxury goods and you have five or six visual styles you want to rotate through: editorial fashion, lifestyle, product close-up, campaign, seasonal. Find the best reference ad for each style. Reverse-engineer each one into a prompt. Store them.

Now you have a prompt library. Every time you need a new product image, you pull the style prompt, swap in your product, and generate.

Your visual brand stays consistent. Your production time drops. And every style in your library was originally designed by some of the best creative teams in the world.

What This Looks Like in Practice

The same approach works beyond product ads. Say you need space or interior visuals that match a specific editorial look.

Find a few photography references that have the mood you want. Reverse-engineer each into a camera language prompt. Then adapt each for different spaces or products.

Instead of paying for a full professional photoshoot, you can build a library of photorealistic images in the right style.

That’s what this technique can do for any product or space business.

Do This This Week

Pick one ad you already saved because it looks expensive. Not 20. One.

Paste it into Gemini. Ask for the camera language. Change one noun. Generate 3 versions. Keep the one that still looks like the reference.

Then store the prompt in a note you can reuse. If this turns into a rabbit hole, stop before it becomes busywork. One style is enough for week one.

If you want the rest of the week to stay clean while you test this, keep the plan simple and protect one block for the experiment.

Try It Today

Find an ad or product photo you love. Paste it into Gemini. Ask it to reverse-engineer the image into a reusable prompt.

Take the prompt. Change one element. See what comes out.

It takes about 10 minutes. And the first time you do it, it’s a little bit mind-bending.

If you want a simple first system for keeping experiments like this from taking over the week, start with Focus Filter.