The first time an AI influencer gets a sponsored post, there's a moment every creator runs into. You ask for "her wearing the new ring from Brand X", and the model happily draws a ring. It looks great. It's just not Brand X's ring.
We hit this ourselves while building ads for a small wearable device. The written idea named the product correctly, the storyboard looked polished, and every frame showed a gadget that had never existed: the wrong shape, the wrong colour, a glowing light the real device doesn't have. A brand would reject that ad in seconds, and they'd be right to.
Here's what we changed to fix it, and what you can and can't say about the product.
Real photos of the product go in before anything is drawn.
Why AI models invent your product
Image and video models learn what things look like from enormous numbers of pictures. A famous product that appears in millions of photos, like a well-known phone or a classic sneaker, is something they can often draw from its name alone. A product that's new, niche, or from a small brand is different. The model may know the name as text, but it has no real picture of it, so it fills the gap with something plausible.
The invented product fits the category, so it doesn't look wrong until you put the frame next to the real thing.
Three habits make it worse:
Writing only the product name. "The Aurora 2 smart ring" tells the model almost nothing it can draw.
Describing the product from the idea, not from the product. If the first brief says "a sleek white device" and the real one is black, every later step copies the mistake.
A good AI UGC ad is decided before anything is generated: in about 60 words of script. Here is the five-beat structure our AI influencers' producer writes to, how many words fit in 20 to 30 seconds, and the mistakes that make a script sound like an ad.
Going straight to video. A video model that starts from an invented product has no chance of correcting it.
Step 1: Collect real photos of the product
Before anything is drawn, get good photos of the actual product. The best source is the brand itself: most companies have a press kit or media page with clean product shots, and if they're paying you for the post, ask them directly. Their product pages are the next best place.
What helps the model most:
A clear front view on a plain background.
One or two other angles, so the model understands its shape in 3D.
A photo of it in use: worn, held or on a table, so the scale is clear.
No busy graphics. Banner images with headlines, price tags and collages confuse the model, which may try to copy the text too.
Two to four good photos beat ten mixed ones. Also check you're allowed to use them. Product photos are usually the brand's copyright, so for a paid post, use what the brand supplies or approves.
Step 2: Describe it in plain physical words
Reference photos do most of the work, but a short written description keeps the model honest when the product is small in the frame. Write what a person would see, not what the marketing says:
A matte black disc about 4 cm across, thin, with a small silver logo in the centre, worn on a black strap on the left wrist.
Size, colour, material, shape, where the logo sits and how it's used. Leave out "sleek" and "premium": the model can't draw them, and they push it towards a generic gadget. If models keep adding something the product doesn't have, like glowing lights, don't write "no lights" (naming it can draw it); describe the real surface instead: "a matte, unlit black surface".
Step 3: Put the product in a story, not a catalogue
A product photo with a face next to it doesn't do much. The posts that work make the product part of a small moment in the character's life, built on how it really works (see how to research a product before an influencer ad). Three patterns that work in pictures, with no speech:
Before and after. The problem in the first shot, the product in the middle, the relief at the end.
The quiet hero. A busy, chaotic scene where the product is the one thing that works.
The routine. The product as part of something she does every morning, shown in three or four quick cuts.
Keep the product clearly visible in at least the key shot. For more on telling stories in pictures, see How to Write AI Video Prompts.
Step 4: Draw the frames first, then check them against the photo
This is the step that fixed our problem. Before making any video, draw the still frames: the opening shot, the payoff, and any shot where the product is close up. Give the image model the product photos as references alongside the character's own references, so it has both the face and the product to copy.
Then compare each frame with the real product photo, side by side:
Shape and proportions: is it the right size compared with her hand or face?
Colour and finish: matte or glossy, the exact shade.
Logo: in the right place, and not misspelled. Models still garble small text, so a small or partial logo is often safer than a large one.
How it's used: worn on the correct wrist, held the right way up, plugged in where it should be.
Redraw anything that's off. A frame costs cents; a video built on a wrong frame costs dollars. Our AI video cost calculator shows how fast re-shoots add up.
Check every frame against the real product before you make the video.
Step 5: Keep the product right once it moves
Video models are better at motion than at details. Over a few seconds a product can slowly change shape, lose its logo, or swap colours. What helps:
Start from your approved frame. Use image-to-video, not text-to-video, so the first frame already has the right product in it.
End on an approved frame too. Models that accept an end frame (Kling does) will land the last shot on a picture you've checked, which is usually the product shot.
Keep product close-ups short. A one- or two-second close-up holds its shape much better than a slow eight-second orbit around it.
Describe the action, not the product, in the cut list. "She taps the disc on her wrist and smiles" works better than repeating the product description in every cut.
Our free AI video prompt builder lays out a cut list in the format Kling and Wan follow, with an optional end frame.
Step 6: Be careful with claims and labels
An AI character hasn't used the product. In the US, the FTC's Endorsement Guides were updated in 2023, and the updated definition of an endorser is broad enough to cover virtual and AI-generated influencers. Endorsements have to reflect honest opinions and experience, and an AI character can't have experience. So avoid lines like "I've worn this every day for a month" or "it fixed my sleep". Show the product, describe what it does in the brand's own approved terms, and let the brand stand behind its claims.
Claims need backing. Anything about results ("improves your recovery", "lasts 48 hours") should come from the brand, who must be able to support it. Don't let the AI write new claims for you.
Don't show results the product can't deliver. A flawless before-and-after drawn by an AI model is a picture, not evidence. If the "after" is invented, don't present it as what the product does.
Label both things. A sponsored post from an AI influencer needs two disclosures: that it's an ad, and that the content is AI-generated. On Instagram, the paid partnership label (available to Creator and Business accounts) plus a clear "ad" in the caption covers the first, and the platform's AI label covers the second. We cover the rules in detail in Do AI Influencers Need to Disclose?. This isn't legal advice; for a big campaign, check with the brand's legal team.
A checklist before you send it to the brand
Two to four real product photos, from or approved by the brand
A plain physical description: size, colour, finish, logo, how it's used
A small story where the product matters, told in pictures
Every frame drawn with the product photos as references, and checked side by side
Video made from approved start and end frames, with short product close-ups
No invented experiences or new claims, and both labels on the post
When you ask your influencer for an ad, Influgen looks the product up on the web first and saves real photos of it before any idea is written. Her team's ad manager then writes a short ad brief: what the product does, the one message the ad must land, the moment that shows it working, and what it must never claim. Her team makes the ad inside that brief, the product photos go into every frame that shows it, and the reviewer that checks the frames redraws any that get the product wrong. When you open the post suggestion, you see the ad brief, the product photos it was made from, the storyboard and the video plan (model, length and estimated credits) before you approve anything. Every post goes out with each platform's AI label.
Most AI ads give themselves away through the sound before the picture: dead air, a voice with no room around it, music fighting the words. Here are the seven fixes we now build into every talking ad, with the exact settings we use.
Most weak product ads aren't badly filmed. They're made by someone who never learned how the product works. Here is the research good creators and agencies do before writing a single line, and what we learned building it into our own pipeline.
