CBD pack shots are identity holds; health claims are a different legal file
Seedance the bottle. Do not put a talking health claim on a generated mouth.
Seedance the bottle. Do not put a talking health claim on a generated mouth. The pack is identity. The claim is a legal file. A model that holds the label is not a model that is allowed to say what the label is not allowed to say.
Photograph the unit. Animate the photograph. Keep speech, if you use it, on a person who exists and a script counsel already signed.
Hold the bottle, not a speaker
The buyer needs the cap, the dropper, the exact print on the wrap. Text-to-video will invent a prettier bottle with melted type. That is a failed SKU.
Image-to-video keeps frame one as the file you sent. Seedance 2.5 is the family for that still-to-motion row: a locked plate, four to thirty seconds in one pass, native audio, 720p on the image-to-video endpoint, optional last frame when the shot has to land on a second approved angle — front of pack to the ingredient panel. Prompt only the camera. A slow push, a turntable, a window that was already in the still. Do not prompt "a wellness coach explaining why this oil changed her sleep."
AI product video is the reference-to-video path when the same bottle has to appear in a room you do not have a photo of. Two or three clean angles are the contract. Fine print is the first thing a model loses — check every take at full size.
Health claims do not belong on a generated mouth
A talking head that diagnoses, treats, or promises an outcome is not a pack shot with audio on. It is a claim, delivered by a face. If the face does not exist, you have a testimonial you did not collect. If the face exists and you generated a new performance, you have a replica. If the mouth is reciting a health line the label does not carry, you have a different problem again.
The fake review rule does not care that the speaker was sampled. Fabricated proof is fabricated proof. YouTube's line on AI presenters and sensitive subjects still names an AI "doctor" giving health advice as the shape to avoid. A generated wellness coach on a CBD SKU is that shape wearing a dropper.
Do not lipsync a claim onto a stock face. Do not Avatar a clinician. Overlay type you wrote, on a plate of the real pack, or film a real person with a replica-aware grant. A likeness release that covers AI generation is the document if someone will be replicated. Most CBD ads should never open it. Show the bottle.
Disclosure is a second file
A paid integration still needs a clear ad disclosure. An AI info tag does not tell the viewer they are being sold to. Layer them. Do not hide the commercial fact behind the synthetic fact. That split is the FTC ad disclosure versus the AI label.
On-screen "Ad" or "Paid partnership with …", early, readable muted. Platform branded-content toggle as well when you have one. Neither replaces approved labelling. Model names, milligrams, and "these statements have not been evaluated…" belong in a text overlay you control. Video models do not typeset a Supplement Facts panel.
The clip can show the object. It cannot invent an experience in a mouth that does not exist.
FAQ
Can the bottle have native audio if Seedance 2.5 image-to-video has audio on?
Yes — room tone, a light scrape of glass, a score-shaped bed. That is not a health claim. Quoted dialogue in the prompt is how a locked pack becomes a talking endorsement. If a line has to be said, a real person says it, on a script you already have.
Is a generated "customer" fine if I disclose it is AI?
No. Disclosure does not make an invented experience true. Do not generate a buyer who does not exist. The FTC ad disclosure is about the paid connection, not a licence to fabricate proof.
Can I put the claim in captions instead of on a mouth?
Captions of a claim you are not allowed to make are still the claim. Type is not a loophole. Keep on-screen copy to what the label already says, in words you wrote.
Why not text-to-video a lifestyle kitchen and composite the SKU later?
Because T2V will invent a second bottle, or a speaker, or both. Start from the pack still. If you need a room, shoot an empty plate or generate one with no product in it, then hold the SKU on I2V. Do not ask a video model to invent the mouth the still was meant to avoid.