We let an AI ad generator run a back-to-school drop — five ad sizes, zero resizing

Every small clothing brand I've talked to does the same thing each season. They design one good ad, and then somebody, usually the founder at midnight, manually rebuilds it into a square, a story, a feed post, a banner, a wide homepage strip. Five files that all say the same thing. That resizing work is the actual cost of a launch, more than the design itself.

So we tried to delete it — with an AI ad generator. We took a store that sells premium polo, tees and sweaters, pasted the URL into On Brand Ads, and let the AI produce the whole back-to-school ad set. No brief. No copywriter. No designer rebuilding sizes at 1am.

What the AI ad generator pulled out of the site

Roughly thirty seconds after pasting the URL, the AI had read the store and locked a brand kit: the navy-and-off-white palette, the wordmark, the type, and the product facts worth keeping visible (100% Pima cotton, classic straight fit, men's and kids'). That kit is the part I'd actually pay for. You build it once and every seasonal drop after this inherits it, so the second campaign is cheaper than the first.

I was skeptical the AI would pick a usable headline. It landed on "Sharp-Edged Classics, Ready to Wear" with a "Buy 2 Classics, Get 15% Off" bundle as the offer. That's a better line than most founders write for themselves, and more importantly it's a bundle mechanic, which is the right move for back-to-school where parents are buying multiples anyway. You can edit a word or have the ad generator retry before any image gets made.

The five ad sizes, generated at once

Here's where the resizing tax is supposed to vanish. Same brand, same message, five native sizes — each one generated as its own composition, not cropped from a master file.

The square is the one that has to work hardest. Feed posts, product cards, retargeting, all square, all one glance. The AI fit the headline, four product facts, the offer and a "Classics Bundle" checkout call into a single frame without feeling stuffed, which is the thing that usually goes wrong when a human compresses a landscape ad down to 1:1.

The 4:5 earns a bit more vertical real estate in-feed, so the generator gave it a second beat, the "Outfit Ready, Grab & Go" bag, on top of the discount.

Stories are where most teams quietly fail. They shrink a landscape design into 9:16 and ship dead space top and bottom. This one is composed for the frame.

And the two wides, for banner inventory and the homepage hero:

What I'd actually do with this

A few opinions, because the AI output isn't the whole job.

I'd ship the story and the 4:5 first. Stories and feed are where back-to-school attention actually is; the wide homepage strip can wait a day. Most teams do the opposite and burn the morning on the banner nobody sees.

I'd also have the ad generator produce a second square from the same context and A/B it. We did:

Same campaign, different composition, same brand kit. That costs you almost nothing with an AI generator, and it's the difference between "we ran an ad" and "we ran a test."

The one thing the AI does not do for you: pick which product to lead with. It chose the Pima-cotton polo as the anchor, which is defensible, but if your margin is in the sweaters, lead with the sweaters. The brand read gives you the options; you still make the merchandising call.

Where it lands

The resizing work, the part that kills launch nights, is genuinely gone. You paste a URL, the AI ad generator reads the brand, you pick a message, you get five ad sizes, you keep what works. The strategy is still yours. That split — the AI doing the mechanical 80 percent and you owning the 20 that actually matters — is the whole pitch of generating ads this way.