Picture this: It's 2am. You need 15 variations of your hero product shot for tomorrow's Meta campaign. Your designer is asleep (reasonable), your photographer costs $200/hour (unreasonable), and your current product photos look like they were taken in a Walgreens parking lot.

Enter Nano Banana 🍌 Gemini 2.5 Flash Image.

Here’s what I discovered after spending 10+ hours using Gemini 2.5 Flash Image, on top of the hours we spent integrating it into our product at Raleon.

Here's the thing about AI image models: they've all sucked at the stuff that actually matters for DTC. Every model fails some or all of the same 6 tests:

ChatGPT fails on all parts except for #6. It’s creative in a savant sort of way. Sometimes you get avantgarde, sometimes you get useful.

Flux 1 Kontext wins on #1, #2, and #4 but fails on the rest.

Ideogram Character win on #1, #2, and #6 but is limited to people.

Flash Image literally hits every single one. With one important exception. It’s better than every other model at text, but it’s still limited and writes like a 3 year old 50% of the time.

So Flash wins on paper. But does it actually work when you're staring at a blank Meta Ads Manager at 2am, questioning your life choices and wondering if your college roommate who became a dentist had the right idea all along?

When it comes to using Flash Image, there’s a new feature that is the difference between getting amazing results or an extra arm in your image: the ability to “build” on a scene.

What I mean by building on a scene is you can start with a scene, character, or product, and then step-by-step add to it.

Gemini calls this feature “Multi-turn” edits.

In fact, with Flash Image you want to avoid “single turn” edits. That’s where you stuff a bunch of changes into one prompt, like “zoom out, put it in a fancy bathroom, add such and such text, and make them happy and dancing on one foot.”

Let's be honest, you didn't read this far to learn about image generation theory. You want to know if this thing can actually pump out ads that don't look like they were made by an intern with a Canva addiction.

The below is an ad I created using multi-turn edits in 3 steps.

Step 1: Reference image

I started with a product shot of the woman. In this case, I pulled an image from Jones Road Beauty’s website. I uploaded it to the chat, and my prompt was:

Step 2: I added a reference product image

I then grabbed a product image from Jones Road Beauty’s website. I uploaded it to the chat, and my prompt was:

Step 3: Make it feel like an ad

You can be much more specific than I was, but now that I had the creative looking how I wanted, it was time to use it to layer on the ad features.

I added the last bit about keeping things the same as I was not sure how much Flash Image would try to change it based on “turning it into a meta static ad”.

The result is in less than 5 minutes I created the ad above. That’s about 90% less time it would normally take someone to create a net new ad, unless they’re using solid templates.

The challenge of working through what we just did above is it takes a lot of time. Especially to get to different versions. It’s a lot of typing, uploading, and iterating.

That’s why I’m going to show you how to create your own ad creator that uses Flash Image.

After about 60 seconds, it should be done building, giving you something like this

With all that, in about 60 seconds, you’ve got your own Flash-powered ad factory. I’ve been using it this week, and actually generated a few Raleon ads with it that we’re testing.

And that's it for this week's edition. Flash just became the creative designer you can't afford to ignore. Now stop reading newsletters and go generate some ads!