Back to the list

Reviewing AI Product Shots

I check the product and action before the atmosphere. A practical review table for creative teams.

When I review an AI product shot, I start with the sound off. Without the music doing some of the persuading, I can concentrate on the product, the contact between a hand and an object, and whether the action makes sense.

A beautiful shot can still show the wrong product. I treat product accuracy as a separate check; good lighting cannot compensate for an extra button or a missing port.

Establish the correct product

I want the approved product photographs, packaging artwork and feature information beside the review. References should cover the angles the camera will actually see. A single polished front view leaves too much unanswered when the product turns around.

The version matters too. Older packaging or a different model can remain perfectly consistent throughout a film and still be wrong for this job.

For a hypothetical example, imagine a desktop speaker with a front dial and two rear ports. If a third port appears during a turn, I would send the shot back. This is an illustration, not a client case.

Before review, I would identify the features that must stay fixed: silhouette, component count, logo position and operation. Lighting may change how a finish looks. It should not change how the product is built.

Watch the movement, then inspect the frames

First I watch at normal speed. Does the action finish? Can I see what the product does? Is an important part hidden at the wrong moment?

Then I inspect the transitions: a new angle, hand contact, an object emerging from behind something, or the cut into the next shot. Checking only the first and last frames can miss a brief deformation in the middle. Checking only still frames can hide an action that is difficult to understand in motion.

Check What I need to establish Example of a problem
Appearance Shape, components and markings match the approved references A dial becomes a button
Action Contact and movement make sense A lid opens before the hand touches it
Function The image stays within the confirmed product claims An effect suggests an unsupported feature
Continuity The product state carries between shots A disconnected cable reappears
Delivery Text, sound and framing work in the final version A vertical crop removes the operation being explained

This is a review aid, to be adapted to the agreed brief. It is not a universal scoring system for AI models.

Choose the right repair

A misplaced caption and a changing product silhouette need different responses. A text layer may be adjustable in post. Repeated structural changes may call for a simpler movement, a better reference angle, live action, an accurate 3D asset or compositing.

Generated imagery can help explain a concept. It does not become evidence of battery life, strength or real performance simply because the demonstration looks convincing. Those claims need the appropriate product information and evidence.

Approve an actual deliverable

An approved textless landscape master does not automatically approve the vertical or translated versions. Each can introduce new cropping, timing or readability problems.

I want an approval to identify the file, version and checks it covers. “Looks good, polish the details” leaves the next person guessing.

For the relationship between shots, see AI Shot Continuity. For more useful revision notes, see Giving Feedback on AI Ads.

WeChat

LEO CHAN · WeChat QR code

Scan with WeChat, or copy my ID to add me.

Open original QR image