Your creativity is the only limit.

How to use the most recent AI models to create everything from image ads to complex videos.

If you’ve ever run a campaign on Google Ads or Facebook and Instagram, the biggest obstacle before launching it was probably creating the ads—the creative part.

I bet most of you aren’t designers and don’t enjoy designing, taking pictures, or shooting videos.

I’ve witnessed the evolution of AI since its modern renaissance and worked with models far less advanced than what’s now freely available. Back then, it was hard to get those models to produce anything remotely usable. Especially on the visual side, there was no way to create something from scratch that could actually be used in a real campaign.

One of the biggest issues was always the text. Those models couldn’t properly generate text within images—you’d end up with some unreadable, hieroglyphic-looking mess. If you’ve ever generated an image with AI, you know exactly what I’m talking about.

But things have changed. New models like FLUX, Recraft, and GPT-4o can now reproduce long text accurately. This is a game changer, and we’re going to use these models to craft an image ad—and even a video.

Let’s say we want to create an ad for a Nutribullet smoothie maker. All we have is some text and an image of the product:

And here’s the text we’re working with:

Power your day. 900 watts of power in this compact powerhouse.
Essential design. Simple, intuitive design makes this Bullet® a device you’ll use every day.
Nutrition in no time. Did you know most smoothies can be made in under 60 seconds?

As usual, we’ll head to the Meta Ad Library for inspiration. And, like almost every time, I’ll be using AG1 as a reference point. (I’m not a fan, quite the opposite actually — but they’re great at marketing and spend a lot of money testing ads.)

This is our source ad:

Now we head to ChatGPT and prompt something like this:

Take the image ag1.jpg (source ad) and create a version of it for Nutribullet, using the nutri.jpg image (the Nutribullet product photo).
Change the text to match the features of Nutribullet:
Power your day. 900 watts of power in this compact powerhouse.
Essential design. Simple, intuitive design makes this Bullet® a device you’ll use every day.
Nutrition in no time. Did you know most smoothies can be made in under 60 seconds?
Remove the testimonial section and add a strong headline.

In less than a minute, I got this:

No typos, perfect integration of the product. The spacing isn’t perfect, but honestly, I didn’t do anything besides asking ChatGPT to generate it. I even got a portrait version:

This is crazy!

Imagine how many variations we can create, how many tests we can run at zero cost. We can start from scratch, use another ad as a seed like we just did, or pull different pieces together and ask ChatGPT to assemble them. Sure, we don’t have full control — but do we really need it when we have the freedom to generate and experiment like this? And if needed, we can always import these images into an editing tool and fine-tune them however we want.

Just imagine how many variations we can create—how many tests we can run—at almost zero cost. Sure, we don’t have full control, but do we really need it when we have this kind of creative freedom to generate and experiment?

And now comes probably the most wow moment (to be honest, I was already impressed with the image generation): we can turn this ad into a video.

Video generation is still a bit behind compared to images, but with some experience and finesse, you can go surprisingly far—getting results we couldn’t have imagined just a year ago. I find it incredibly inspiring.

For this, I’m using a model called Kling (through Replicate.com), and I’m inputting the ad we just generated. I simply ask the model to transform the fruits and vegetables into a pink smoothie. Here’s the prompt:

"The camera pans in and the vegetables and fruits inside the glass part of the blender become a pink smoothie."

With video models, it’s important to spell out even the obvious—like making sure the smoothie appears only in the upper part of the blender.

A couple of minutes later, I get this:

I found it remarkably good! And if you want, you can even go wild—ask the model to add more details or change the camera movement. It’s a trial-and-error process. Video models can be a bit wild to tame, and you might need some time to get solid results—especially if you go crazy with your prompts.

If you want to see the whole process in action, I’ll record a video and share it on my YouTube channel. Don’t forget to subscribe: https://www.youtube.com/@theadsdude

We are living in a fantastic era.

Creativity is now the only limit. AI allows us to generate and iterate in ways that required entire teams and massive budgets just five years ago. You literally have zero excuses now when it comes to creating imagery and videos.

Feel free to get in touch if you want to learn more about my creative process. I’m happy to share all the secrets!

That’s all for today! Feel free to share this newsletter with anyone who might be interested. If you bring in 5 new subscribers, you’ll get a copy of my Google Ads book (a $39 value)!

See you soon!

Francesco (The Ads Dude) - adsformakers.com