AI Creative Testing
Nano Banana — Gemini's image generation and editing model — can drop your product into any scene, keep it pixel-consistent, and produce a dozen ad variations from one photo. Which one stops the scroll? That's not a question an image model can answer about itself.
The thing Nano Banana does that older image models couldn't: keep your actual product looking like your actual product across edits. Change the background, relight the scene, move the bottle to a beach at sunset — the label stays right. That consistency makes it genuinely usable for ad creative, not just concepts. Combined with conversational editing ("make the background warmer, push the product left"), a designer's day of comps becomes ten minutes of prompting. The result is the same everywhere though: teams drowning in variations with no way to rank them.
Static ads live or die on instant categorization — a viewer decides in well under a second whether your image is worth a glance. Nano Banana's outputs are aesthetically strong by default, but aesthetics aren't salience. A cluttered scene, text competing with the product, a background that swallows the subject: these kill attention regardless of render quality. It's the pattern behind most banner design failures, and color and contrast choices the model makes for beauty aren't always the ones that make an ad legible at feed speed.
PreTestAds scores static images by converting them to a 10-second visual stimulus and running the same fMRI-trained attention model used for video — you get a percentile score against top-performing ads plus a full report. The loop with Nano Banana is unusually tight because editing is conversational: score the image, read what's failing, then tell Nano Banana exactly what to change — "the aerial background is chaotic, replace with a clean ground-level shot and add a dark vignette behind the price." Re-score the edit. Our KAYAK case study took a static from a 5/100 to an 88/100 in three rounds exactly this way.
The highest-leverage use: generate the same product in ten different scenes and score the batch. You'll learn which contexts make your product salient — knowledge that transfers to every future ad. Pair with Midjourney statics to test different aesthetic directions, or step up to the full AI testing workflow when you're ready to add motion with Veo.
Upload your generated statics and find the one that stops the scroll — first analysis free.
Test Your Ad Images