How to make product ads with AI
A product ad is three decisions in one image: what grabs attention in the first instant, what the product is, and where the words go. AI makes the production cheap enough that you can decide them by testing rather than by arguing.
Design the first instant
Feed ads get judged before they are read. The subject should be doing something in frame — mid-motion, mid-use, mid-pour — because a static product on an empty background gives the thumb nothing to catch.
Write that into the brief: not "product on marble" but "pouring, mid-stream, condensation on the glass, dark kitchen out of focus behind".
Build for the placement, not the format
9:16 story, 4:5 feed and 1:1 tile each have a different centre of attention and a different UI overlay. Generating one image and cropping it three ways is the most common reason an ad underperforms.
Name the placement in the brief so the composition is designed for it, and keep the safe area clear of anything you cannot afford to lose.
Leave the copy space on purpose
Headline, price, badge, disclaimer — each one needs somewhere to land that does not fight the image. Ask for negative space in a named region and the frame gets composed around it.
This is why briefs that only describe the product usually produce images that cannot carry an ad at all.
Static and short video
Static ads test the angle; short video tests the motion. Video mode renders one continuous shot per send with a seconds selector, which is the shape that loops well on TikTok and Reels.
Choose the video model by the motion you need — fast effects, realistic physics, or lifelike human movement — rather than accepting the default and hoping.
Use a person without losing the thread
UGC-style ads outperform polished ones because they look found rather than made. The catch is continuity: without a defined character, shot two features a different person and the set falls apart.
Create the character once, then generate every shot with the same person, room and framing. That consistency is the whole trick.
Get the winner out
Keep the variants on one canvas until you pick. Then push the winner to Meta Ads, Instagram or TikTok directly — all connected — instead of downloading, re-uploading and re-tagging it.
Testing beats opinion, but only if the losing variants are cheap. That is what the brief-to-batch flow is for.
Questions people ask
How many ad variants should I make before testing?
Four, on one axis. Angle, setting, framing or person — change one thing per batch, or the results tell you nothing. Four also fits on one canvas, which is what makes the comparison fair.
Can AI-generated ads look too obviously AI?
They usually look generated when the image is trying too hard — perfect symmetry, floating objects, five fingers. Briefs that ask for an in-the-moment frame with copy space do noticeably better than briefs that ask for a poster.
What size should a product ad be?
Ask for the placement by name — 9:16 for stories and Reels, 4:5 for feed, 1:1 for square tiles — and the composition is designed for it rather than cropped into it.
Make the batch, then let the platform pick
One brief, four variants, Meta Ads and TikTok connected.