Hands, products, and the contact problem

01Why is contact harder than anatomy?

Anatomy is one object. Contact is two, and the second one is yours. Fingers stopped being a joke faster than anyone expected because the world has photographed hands endlessly. Nobody has photographed your pack, so the two have to agree about size, shape, pressure and shadow while the model is guessing at half of what it is holding.

The giveaway is that generated contact is one-way. Real contact deforms both parties: a fingertip flattens against a hard pack, skin goes pale at the pressure points, a soft tube dents. In generated footage the hand does all the work and the object does none. Look only at the pixels where the two meet and you will be right about the shot inside a second.

02Your product gets improvised, and the size goes first

A model has never seen your product. What it has seen is the category, so it draws the average member of that category, at the average size, held the average way. The failure is rarely the object itself. It is the scale. Serums arrive the size of a wine bottle. A compact arrives the size of a hand mirror. Everybody on the review call spots the color first and the size never.

Four of our own frames, and what each one is asking a viewer to believe

Ours, made this way

A wrap grip on a hard, wet, transparent object. Four difficult things at once.Contact against skin, where the device has to press and the cheek has to give.Skin on skin, wet. The hardest contact on this page and the least forgiving.Contact with no hand in it. Often the right answer, and always the cheaper one.

None of these is a failure. They are the frames where a failure would have been obvious, which is why they were the ones watched hardest before they shipped.

We started writing sizes down after a hotel film rendered a stone entry crest at roughly eight feet when the real one is about five. Every shot naming a physical thing now carries its measured size beside the person's height, marked measured, owner-stated or assumed, and scale anchoring earns a piece of its own.

03What if the product doesn't fit in the character's hand?

Then stop trying to make it fit and change the shot. A product that reads wrong in a hand is a framing problem with five known answers, and four of them are cheaper than another generation. The order we try them in: cut away from the hand, frame the hand out, drop a real photograph of the pack in and move it as a card, composite a real hand, shoot the thing. What a still is allowed to do once it starts moving is set out in the physics test battery.

04The real-hand insert, and the shadow that gives it away

This is the route people skip because it sounds like a visual effects job. It is twenty seconds of phone footage, a mask, and one shadow built correctly. The shadow is where almost every attempt falls down, for a single reason most people never notice they chose.

05How do you make AI UGC look real?

Make the two hardest things in the frame real: the contact, and the voice. Everything else can be generated without anybody noticing. In practice that means a real photograph wherever your label has to be read, a real hand wherever the product is handled, and an unscripted delivery with gaps in it, because the creator register is set by audio far more than by picture.

One thing to steal from that scale slot, because it generalizes past hands. State the relationship, not the number. Reads about chest-to-eye height beside her produces a better result than five feet tall, because a model has no ruler and does have a very good idea of what a person is. The same instinct governs the rest of the sheet: describe what the frame should contain, not what the object measures. The failure modes behind all of this are cataloged in the frame-by-frame atlas of why AI ads look fake.

06When should you stop and shoot twenty seconds yourself?

The moment you have paid for the same shot twice. A second failed generation of a grip is not bad luck, it is the same blind spot returning, and a third will return it again. Twenty seconds on a phone, in one window's light, costs less than the credits already spent and produces something no prompt can: your actual product, at its actual size, in a real hand. If it turns out you need a face and a voice as well as a hand, that is a larger decision, and when you should still hire a human creator is the honest version of it.

TWENTY SECONDS. ONE WINDOW. NO CREW.

Phone propped on a stack of books, lens level with the product.
One window on one side. Turn every other light off.
Wipe the product first. Fingerprints are the whole shot.

00-04   Hand enters. Pick the product up.
04-07   Hold still. Three full seconds. Do not adjust.
07-15   The action: peel, press, pour, twist, open.
15-20   Set it down. Hand leaves frame. Keep rolling.

Shoot it twice. Send both files.
No captions, no music, no filter, no crop.

And the footage the brand already has

A client's own clips are the fastest way to solve contact and the fastest way to make an ad that looks like their last one. Borrowed material is never used raw here. It gets recropped, regraded and recomposed, and their captions and phone audio never survive. Borrowed footage is punctuation, never the spine, because the transformation is the only part anybody is paying for. If every hard shot is answered by a clip they already own, you have edited their camera roll.

07Hands are where our re-roll rate is worst

We have no clever answer for that. What we have is a rule that stops the bleeding: a bad take is re-rolled, never repaired, and every re-roll is somebody's decision before it is anybody's spend. Nursing a broken grip through four rounds of fixes is how a cheap shot becomes an expensive one. The economics are laid out in re-roll, never repair.

The other honest limit: no film in our reference set carries a hook rate or a return on spend, so we can tell you which contact shots survive a human watching them and not that a better grip sold anything. Some categories are simply easier on this axis, and what kinds of product AI UGC actually works for is a better question than how to force the hard ones.

Some products are held in ways nobody has ever filmed, and for those the honest answer is a camera and an afternoon, not a better prompt. Knowing which of your shots is that shot is most of the skill. Everything else on this page is refusing to ask a machine about an object it has never touched.