The Tell Is in the Teeth: How to Spot AI Brand Imagery Before Your Audience Does
Photo: SMU Central University Libraries, No restrictions, via Wikimedia Commons
The pitch deck arrives with forty frames. The lighting is gorgeous. The model has great bone structure. The client says yes in eleven minutes. Then the ad runs and someone in the replies says the word "AI" and the whole thing collapses into a debate about ethics instead of a conversation about the brand.
That's the trade nobody puts on the estimate. A synthetic image doesn't just cost you a photographer's day rate. It costs you the thing you were actually buying: proof that somebody went out into the world and looked at something.
The Prompt Has Never Touched a Mud Puddle
Here's what I keep coming back to. A photograph is a record of a decision made under constraints. The sun was in the wrong place, the permit ran out at four, the model's shoe was wet, the gaffer held a bounce card with his hands because the stand broke. All of that pressure leaves marks. Those marks are the reason the image feels alive, and they're the first thing a generator deletes, because a generator has never been cold.
So the argument isn't really real versus fake. It's pressured versus unpressured. A prompt produces an image with no consequences in it. And American audiences in 2026 have gotten extremely fast at sensing an image that cost nothing.
The Tells, Ranked by How Fast They Get You Caught
Teeth and hands. Still. Always. Teeth in synthetic imagery tend to arrive in neat identical rows, evenly toned, with no gum recession, no coffee stain, no chip on the incisor from a bicycle wreck in 1998. Hands will have correct finger counts now and still fail on the small stuff: knuckle wrinkles that don't line up with the grip, a thumbnail that doesn't catch the same light as the rest of the hand.
Light with no source. Three-quarter key, soft fill, a rim that comes from nowhere, and not one practical in frame that explains any of it. Real sets have a visible reason for every highlight. Synthetic sets have a mood.
Fabric that doesn't obey. Denim should be heavy. It should bunch at the crotch and drape off a hip. Generated denim often reads like a printed texture, and the seams will wander around the garment. Tension is physics, and physics is expensive to fake.
The grain sits on top instead of underneath. Real high-ISO noise lives in the shadows and changes with exposure. Faked grain is a filter. It's uniform, it's the same in the highlights as in the dark, and it doesn't respond to the lens at all.
Lens behavior that doesn't add up. Bokeh shapes should match the aperture and the number of blades. Falloff should be consistent across the frame. Reflections in a window should contain the scene you can't see but can infer. This is where generated images fall apart on a big screen and skate by on a phone.
Signage, lettering, and logos. Still cursed. Menu boards, hand-painted truck doors, the license plate, the tattoo. Anything that demands a specific arrangement of marks will eventually produce a letter that doesn't exist.
The uncanny consistency of skin. Real skin is uneven. One cheek is flushed, the nose is peeling, there's a bruise fading on the forearm. Generated skin across a series tends to have the same pore density on every subject, as if the whole cast shares one dermatologist.
The Two-Question Test
When a client asks whether you can just generate the hero shot, run this:
- Does the image need to prove something happened? If the strategy depends on authenticity — a real town, a real worker, a real Tuesday night — synthetic is disqualified on principle, not on quality. You can't prove a document with a drawing.
- Will the viewer ever see it at full size, for more than four seconds? Social cutdowns and internal comps forgive almost everything. A billboard on I-70 does not.
If the answer is no proof needed, small and fast, generate away. If either answer flips, you're not saving money. You're deferring a cost to the day the comment section finds you.
Where Synthetic Actually Earns Its Keep
Be honest about the upside, because there is one. Generative frames are phenomenal for previsualization — showing a skeptical brand team what a hard sidelight will do to a bottle before anyone books a stage. They're useful for backgrounds, for extension of sets you already shot, for cleanup, for variants in a paid social test where the words matter more than the pixels. They're great for the frame you'd never get permission to shoot.
What they can't do is be the reason the campaign is interesting. Bold imagery comes from friction. A generator has no friction. It only has taste, and taste is the least rebellious thing a brand can buy.
Put the Risk Back in the Frame
The uncomfortable reading of all this is that AI imagery is exposing how many campaigns were already hollow. If a brand's photography was already generic, already interchangeable, already styled into beige — the generator didn't ruin it, it just made the emptiness legible overnight.
The defense is the same as it's always been. Shoot the thing that could only have come from that place, that day, with those people. Keep the blown frame. Keep the mud. Keep the kid in the background who wandered into the shot and made it better. Nobody can prompt that back into existence, and that's exactly why it's worth the invoice.