Product accuracy
Product and label accuracy in AI commercials
For the brand manager whose product has a label legal has approved, a logo that cannot move and a shape customers know on sight, and who is about to put it in an AI-made commercial.
The short answer
A video model redraws your product in every frame from your reference images and prompt, so label text, logos, proportions and colors can drift from shot to shot. The fixes, in order: send clean references of every side, start product shots from an approved first frame, keep readable text out of generated shots, set the real logo, label and claims in the edit, end on a pack shot built from a real photo, and check every frame against the real product before legal sees it. An ad that shows a product looking or working unlike the real one is misleading, however it was made.
The cause
Why AI video changes your product
Nothing in an AI shot is pasted in. The model studies your reference images and draws the whole frame again, product included, fitted to new light, a new angle and motion. It can get the light right and the fourth letter of your brand name wrong. Video multiplies the problem: a 6-second shot at 24 frames a second is 144 drawings of your product, and every turn, reflection and hand that touches it is another chance to drift.
The model makers are working on it. Veo 3.1 takes up to three reference images of a product to keep its appearance in a clip, and Google says Gemini Omni Flash renders prompted text correctly and readably. Neither is a guarantee for a regulated label. For stills, the same cause is taken apart in Overs' guide to why AI changes your product.
What drifts
What drifts, and why it matters in an ad
| Detail | How it drifts | Why it matters | The fix |
|---|---|---|---|
| Label text | Letters warp or change; small print turns to noise | A wrong word on a label is a wrong label | Keep small print unreadable in generated shots; composite the real label |
| Logo | Shape, spacing or color shifts | A brand-guideline and trademark problem | Place the real logo file in the edit |
| Shape | The bottle grows taller; the cap changes | Customers know the shape on shelf | References of every side; the dimensions in the prompt |
| Color | The pack shifts under the scene’s light | The brand color is part of the product | Hex codes on the brief; grade back to the reference color |
| Count and size | Nine gummies become eleven; the jar doubles in size | Quantity and size are claims | State the count; a reference of the product in a hand |
| Finish | Matte turns glossy; glass turns to plastic | It misrepresents what the buyer gets | Material words; references shot in similar light |
| Physics | A pour runs backward; a cap unscrews the wrong way | It looks fake, and it can mislead | Keep the action simple on the board; rerun or cut around it |
The fixes
The fixes, in order of payoff
- A reference pack. Every side on a plain background, the label and logo artwork, close-ups of the closure and texture, the product in a hand for scale, the exact colors and dimensions, and a 3D file if one exists.
- An approved first frame for every product shot. The label, logo and shape are checked on a still before any motion is paid for. See first and last frames in AI video.
- No readable text in generated shots. Turn the label a few degrees, soften the focus, or let motion carry it. The words the viewer reads come in step 4.
- Real files in the edit. The logo, the label, supers and any claim are set from the brand's own files, never drawn by the generator.
- A real pack shot. End on the product from a real photo or a 3D render, animated in the edit, so the last image the viewer sees is exact.
- A frame-by-frame check. Someone who knows the product compares every product shot with the real thing, at full size, before the cut goes to legal.
- Legal or MLR review of what the product is shown doing. A generated shot of the product at work is a claim, the same as a line of copy.
Every frame Ruminate X delivers is made with generative AI. There are no film crews, sets or location shoots. In the Ruminate X pipeline the product gets extra generation passes, and the real logo, label and on-screen text are added in the edit from the brand's files. More on how an AI commercial is made, and what else breaks, is on AI commercial production.
The rules
An AI product shot is still a claim
United States. The FTC's Deception Policy Statement (1983) treats an ad as deceptive when it is likely to mislead a reasonable consumer about something material. The Supreme Court applied that to a TV demonstration in FTC v. Colgate-Palmolive (1965): the shaving-cream ad showed sandpaper being shaved, the “sandpaper” was sand on plexiglass, and the Court held the undisclosed mock-up a material deceptive practice. A generated shot that shows a product doing what it cannot do is the same problem in a new medium.
United Kingdom. The ASA ruled in 2021 that beauty filters that exaggerate a product's effect should not be used in ads. Its May 2025 guidance says a disclosure alone will not fix a misleading ad, and that an AI image of a cosmetic's effect that does not match real results is likely misleading even with an AI label. In March 2026 it upheld complaints against an ad for a robotic puppy whose footage, real or AI-generated, exaggerated how lifelike the toy was.
European Union. From August 2, 2026, Article 50 of the AI Act requires those who use AI professionally to disclose deep fakes, and the Act's definition of a deep fake covers objects and places as well as people. Ask your counsel whether a realistic product shot needs a label where you sell.
Checked September 2026. This is not legal advice; your legal, regulatory or MLR team decides for each ad.
The stills
Where Overs comes in
The product stills around a commercial have the same problem, and Overs was built around it. Overs is an AI tool that makes product and campaign photos that match a brand, without a photo shoot. It is the sister product from the same founder, and it was built on the process 100creatives used to make hundreds of photos a week for hundreds of brands.
Product analysis is one of its ten steps. Before a photo is made, the plan names the reference pictures that photo needs, in order, and only those go to the image model; if a needed reference does not exist, Overs skips it instead of guessing with the wrong picture. Among the image models it offers, it lists GPT Image 2.5 Sunburst as the most exact at keeping a product identical while the scene changes. On Pro and Team, a reviewer checks each photo against the references before you see it. Headlines, sublines and buttons go on in a separate typography step, and the clean version is kept, so no render has to spell your claim. The same method, words added in a second pass, is how 100creatives made its gummy supplement ads by hand in 2026, at 3.2 renders per usable photo, the fastest of its 4 campaigns that year.
You see the shot plan and an estimated cost before anything renders. The free plan makes 40 photos a month, and the AI costs a few cents a photo on your own OpenRouter key, with no markup from Overs. Run your product through Overs free at www.overs.studio and check it across a whole set of stills before it goes anywhere near a video model.
Should Ruminate X put your product on screen?
Hire Ruminate X if
- A consumer brand that needs a cinematic ad in which the product looks exactly like the one on shelf.
- A supplement, beauty, pharma or pharmacy brand whose on-screen claims go through legal or MLR review.
- A launch with final packaging and no budget or calendar for a shoot.
Hire someone else if
- The ad demonstrates performance, such as a stain coming out or a blade cutting: film the real product, because a generated demonstration is a claim you cannot back up.
- The product is reflective, transparent or carries small print the viewer must read on screen: a macro shoot or a 3D product render holds it better.
- The packaging is not final: wait for final artwork, because every shot is built on it.
Questions buyers ask
Why does AI video change my product?
Because the model redraws the product in every frame from your reference images and prompt. The product in an AI shot is the model's drawing of it, fitted to the new light, angle and motion, so label text, logo shapes, proportions and colors can drift between shots and within a shot. The more the product moves, turns or is handled, the more chances it has to drift.
How do you keep a product consistent in an AI commercial?
Send references of every side and the label artwork, start each product shot from an approved first frame, use the video model's reference feature, keep readable text out of generated shots, set the real logo, label and claims in the edit from the brand's files, end on a pack shot built from a real photo, and check every frame against the real product before legal review.
Can AI video get a label or logo right?
Sometimes, and it is improving. Google says its Gemini Omni Flash model renders prompted text correctly and readably, as of September 2026. For a label that legal has approved, a studio should still not rely on generated lettering: the real logo and label are set in the edit from the brand's files, and every frame is checked against the real product.
Is it legal to show a product in an AI commercial?
Yes, if the ad shows the product accurately. In the US, the FTC treats a demonstration or depiction that misleads consumers about a product as deceptive, and the Supreme Court upheld that for an undisclosed mock-up in 1965 (FTC v. Colgate-Palmolive). The UK's advertising regulator says an AI image of a product's effect that does not match real results is likely misleading even with an AI label. This is not legal advice.
What should I send an AI studio so my product comes out right?
Photos of every side on a plain background, the label and logo artwork files, close-ups of the logo, closure and any texture, the product in a hand for scale, the exact brand colors, the dimensions, a 3D or CAD file if one exists, and a written list of what must never change, such as cap color, logo position or the number of items in the pack.
How does Ruminate X keep a product accurate?
The product gets extra generation passes, and Ruminate X adds the real logo, label and on-screen text in the edit from the brand's files rather than having the generator draw them. Each shot is checked against its board and generated again when the product drifts. Every frame is made with generative AI, with no crew, set or location shoot.
Send us the product
Send the product, its label artwork, where the ad will run and what legal has to sign off.