Leonardo’s Dwayne Koh on the way AI is evolving as a creative partner

We partnered with OpenAI to launch their strongest and most sophisticated image model to date on Leonardo.With clearer layouts, improved text rendering, and professional image quality from simple prompts, GPT Image 2.0 gives your creative process a running start.

When putting GPT- Image 2.0 to the test, our Head of Creative, Dwayne Koh, began to see something bigger than just technical improvements, he discovered a new workflow for creatives across industries. Seeing how far he could push the aesthetics and art direction of a generated image from a simple prompt inspired him to reimagine the role AI plays in building campaigns.

Here’s what Dwayne discovered:

Maybe the way everyone tests AI is wrong. To this day, the industry seems to evaluate new image models the same old way.

What do the hands look like? How good is it at text rendering? How is it at crowded scenes?

These are useful benchmarks, but they miss the bigger picture. They forget who’s on the other side of the prompt judging the output.

The real test isn’t whether a model can render fingers correctly. It’s whether someone with no prompt engineering skills can type something simple and get something useful. They want visuals that feels like it was crafted by a professional.

Because that’s who will actually use these tools.

So We Tested Something Harder: Advertising

We decided to test the new GPT image model on advertising.

Because advertising is a high bar. It requires:

• Taste
• Art direction
• Audience understanding
• Layout and composition
• A concept that actually lands

And so many decisions need to be made before you can even begin to move into execution.

You can’t brute-force that with technical accuracy alone.

On a personal note, this layout sparked a lot of ideas. I wouldn’t use it directly, I’d simplify it, but it provides a great starting point.

No Tricks. Just Real Prompts.

We deliberately removed any advantage.

No prompt engineering. No hidden techniques.

Just raw briefs —the way a marketer would write them.

What We Saw

The outputs weren’t just visually correct.

They were creatively coherent.

What surprised us most was the detail it added, especially in the Glow Lip ad. It introduced elements we hadn’t considered, like a “viral on TikTok” sticker, then made us realize we should have. It was a smart creative choice designed to build hype.

The model wasn’t just rendering images. It was interpreting briefs, understanding audiences, and making creative decisions behind the scenes. That’s a different thing entirely

The Shift: From Rendering to Training Taste

We’ve been measuring AI on technical outputs.

But what actually matters in creative industries is taste — the invisible layer of decisions that make something feel right.

What’s changing now is that new models are getting better at reflecting taste when we train them.

It’s not up to the model to know what feels right. But it’s getting better and better at producing options for creatives to judge and choose from.

We’re starting to see models that:

• Reflect cultural context
• Suggest aesthetic trade-offs
• Introduce ideas, not just visuals

That’s a different category entirely. Lets push this further into the graphic design world

Now that we have a wide range of letterforms above, let’s use them to generate a full type set, exploring what a cohesive font system could look like , a technique like this could be used to inspire Type Designers

Additional layout explorations with GPT-2, pushed toward bold, wild posting styles to capture the vibe of both a crypto and a soda brand

Which brings us to

The Bigger Unlock: Product Discovery

Most people will stop at using this for ads.

But instead of generating ads for a product, you can generate visuals to discover the product itself.

Designing From Outputs Backwards

Imagine you’re creating a new brand:

• A youth-focused clothing label
• A new lip balm line
• A lifestyle brand that doesn’t exist yet

You generate 20–50 high-quality ad directions.

Not to pick a campaign. But to identify patterns:

• What aesthetic keeps showing up?
• What world does the product live in?
• What kind of people belong to it?

From there, you reverse-engineer:

• The product design
• The brand identity
• The positioning

Inverting the creative process

Traditionally, we go from Idea → Product → Brand → Ads

This tool has the power to flip the process: Outputs → Patterns → Identity → Product

You start with fully formed outputs — and work backwards.

You use abundance to discover clarity.

The models generate the options. We decide what they mean.

Final Thought

We’re inverting the creative process. Instead of beginning with an idea and hoping it lands, we can generate a professional output and work backwards to the product.

That shortens the distance between idea and identity. And unlocks new directions for creators to chose from.

This doesn’t just change how ads are made, it changes how products get invented.