Production
A generated ad can get the light, the mood and the model exactly right and still get the product wrong — a label that reads almost like the real one, a logo drawn slightly off, a pack in a size that does not exist. On a brand shoot that is a retake. In an AI pipeline it is the most common way a finished spot fails review, and in the Philippines it is also a legal problem, not only a creative one.
A wrong product is a misleading ad
The Consumer Act does not care how the frame was made. Article 110 makes it “unlawful for any person to disseminate or to cause the dissemination of any false, deceptive or misleading advertisement” through “print, radio, television, outdoor advertisement or other medium for the purpose of inducing or which is likely to induce directly or indirectly the purchase of consumer products or services”.
The same law treats the label as information the consumer is owed. Article 74 commits the State to “compulsory labeling, and fair packaging to enable the consumer to obtain accurate information as to the nature, quality and quantity of the contents of consumer products”. A generated label that changes the net weight, invents a claim or blurs a warning is not a style choice.
And since 21 September the Ad Standards Council has said the same thing about AI specifically. Advertisers must “ensure that AI-assisted or AI-generated advertising materials are accurate, substantiated where required” under Circular 2026-014, and “reliance on AI shall not excuse or mitigate any violation”. What the ASC now requires covers the disclosure side.
Where generated products usually go wrong
| What drifts | What it looks like on screen |
| Label text | Real words become near-words; ingredients, net weight and warnings turn into plausible gibberish |
| Logo | Letterforms rounded or reweighted, the mark redrawn from memory rather than from the file |
| Pack shape | Proportions stretched, a cap or pump that is not on the real product, a size variant that is not sold |
| Colour | Brand colours shifted by the scene’s lighting, so the pack no longer matches the shelf |
| Claims on pack | Badges, seals and flashes added because they look like packaging, not because they are true |
| Count and state | The wrong number of units, a sealed pack shown open, a product shown doing something it does not do |
Every one of those is a claim about the product. That is why the check has to be run by someone who knows the product, not only someone who knows the shot.
Start from the real product, not a description of it
→ Give the model the actual pack. Google’s documentation for Veo lets you provide up to three images “of a single person, character, or product”, and says the model “preserves the subject’s appearance in the output video”. Use the approved packshot files, not a photo someone took on a desk.
→ Treat the label as artwork, not as content to generate. Where the label has to be legible, composite the real label in post rather than asking the model to write it. Text is where generation is weakest, and it is exactly where the law is strictest.
→ Lock the logo. The logo goes in from the brand’s own file, every time. A redrawn logo is a trademark problem as well as a brand one.
→ Keep a hero frame. One approved, fully accurate frame of the product that every other shot is checked against, the way a model sheet works for a character.
The rule that saves most retakes
If a frame is going to be read — a label, a claim, a price, a warning — it should be composited from real artwork, not generated. Let the model make the world around the product. Let the product be the product.
The accuracy check before it ships
→ One named checker, from the brand side. Someone who knows the current pack, the current claims and what changed last quarter. Not the editor.
→ Frame by frame on anything legible. Label, logo, claims, net contents, warnings. Pause and read them against the real artwork.
→ Every size and every cut. The 6-second cut-down and the 9:16 crop are separate outputs. Check them separately; errors do not stay where you found them.
→ A record of what was checked. Who checked, against which artwork version, on which date. It is the substantiation the ASC expects, and it is what settles a complaint quickly.
General information, not legal advice. Regulated categories — food, drugs, cosmetics and devices — carry their own labelling and advertising rules; check them with your regulatory lead.
Where PUNX fits
PUNX produces AI-powered commercials from Makati for brands including Procter & Gamble, San Miguel, Petron and KFC. The product in the frame comes from the brand’s own files, legible pack copy is composited rather than generated, and the accuracy check sits in the approval chain before anything is exported. Brand-safe AI production covers the rest of the review.
FAQ
Can an AI-generated ad show a product inaccurately?
Not lawfully. The Consumer Act makes it unlawful to disseminate a false, deceptive or misleading advertisement that is likely to induce the purchase of a consumer product, whatever medium or method produced it. The ASC’s AI circular adds that advertisers must ensure AI-generated materials are accurate and that relying on AI does not excuse a violation.
Why do AI video tools get labels and logos wrong?
Generation is weakest at text and at exact reproduction of marks. Label copy tends to become near-words, and logos get redrawn rather than reproduced. The fix is to composite the real label and logo files in post wherever they need to be legible.
How do you keep a product consistent in AI-generated video?
Start from the real packshot files. Google’s documentation for Veo allows up to three reference images of a single product and says the model preserves the subject’s appearance. Keep one approved hero frame and check every other shot against it.
Who should check product accuracy in an AI ad?
Someone from the brand side who knows the current pack and claims, reading every legible frame against the real artwork, in every size and cut-down, with a record of who checked what and when.
Related
AI content production — the service, from brief to delivery
AI disclosure in Philippine ads — what the ASC now requires
Brand-safe AI production — disclosure, consent and the record behind them
Product in the frame?
Send us the packshot files and the brief. We will tell you which shots can be generated and which parts need the real artwork.
Start a project ↗