N95 packs are carton stills; health VO on a generated face is a kill
Photograph the carton. Captions for the count. A talking mask-wearer is a YPP kill.
Photograph the carton. Captions for the count. A talking mask-wearer is a YPP kill.
An N95 pack OEM clip is a still of the box you actually ship: model string, piece count, approval mark, the printed warning that already cleared labelling. Animate that carton. Do not generate a face in a mask and give it a health voiceover. A constructed wearer explaining filtration, fit, or "what you should do" is YouTube's inauthentic-content third bucket — AI personas presenting as human experts on health. The carton is the product. The mouth is a Partner Program kill.
The carton is the SKU
A respirator listing is a specific object. The buyer reads the count, the model, the approval mark. Text-to-video does not. It invents a generic medical box: plausible blue, a cousin logo, a count that is almost yours. That is a different SKU. Procurement notices the panel before they notice your brand.
Photograph the carton under even light. Front of pack large in frame. Lot and count readable at 100%. That still is the contract. If you would not print it on a sell sheet, do not buy motion.
The AI product video generator is built for that contract: reference photographs of the real object, then camera and scene around it. Zoom to 100% and read the small print before you keep a take. Fine text is the first thing a generator loses.
Image-to-video is the single-still door: your upload is frame one. Prompt camera and physics — a slow orbit, a hold on the count — not a second box. Re-describing the carton in the paragraph is how the model draws a second carton. Do not text-to-video "an N95 20-pack rotating on seamless white." Count, lockup, and colourway recast every take.
Captions for the count. Not a mouth.
The number on the box is already printed. Repeat it as type you wrote, after picture lock. Add a text overlay burns one authored line — 20-PACK, 50-PACK, the model string — top, center, or bottom, whole clip. It does not transcribe. It does not listen. Silence, a bed, a cap click: none of that changes the string.
Do not ask a video model to letter the count into the shot. Prompted type on a carton is a misspelled count you cannot edit. Generate a clean plate. Stamp the line you already printed.
A talking generate is not a caption. Overlaying the count on a locked carton is a label. Muted autoplay still needs that count on the file. Do not invent a second panel in the generate and hope the glyphs hold.
A talking mask-wearer is the YPP kill
Synthetic presenters on health and finance are a YPP kill is the policy page. YouTube's third inauthentic-content bucket is AI-generated personas that present as human experts giving advice on health, legal issues, finances, or politics. A diagnosing "doctor" is the Help example. A photoreal mask-wearer walking through fit, filtration, or "what you should do in a wildfire" is the same shape with an N95 in frame.
Labelling the face AI does not leave the bucket. Disclosure is a different control. This policy is about persona plus prescription. A labelled AI clinician is still an AI clinician.
You do not have to abandon respirators. You have to stop using a synthetic expert as the source of the advice. Put a real, named, credentialed person on camera when the video tells a viewer what to do, or do not tell them what to do. Claims stay on the printed carton. Generation can still do the box and the titles. It cannot be the clinician.
Do not generate patients. Do not generate residual limbs. Do not generate fake clinicians. A mascot carton is not this bucket. A photoreal wearer styled as the person who knows respirators is.
Clinic /for is a facility. This is a pack.
Urgent care clinics sell a wait comparison: what the site treats, what it sends onward, a walkthrough of a real waiting area. That page is a facility plate plus a named provider when advice is in the brief. It is not an OEM carton. Copying the clinic playbook onto a 20-pack — generated waiting room, generated patient, generated "doctor" in an N95 — is how a pack listing becomes a health persona.
OEM content is the box on a table, the count overlay, the model string that matches the spec sheet. Seasonal copy can sit on that plate. A generated clinic cannot. Pause on the panel. Spec-sheet carton: this page. Real waiting room and real clinician: /for. No to both: you used text-to-video, or you put a mouth on a mask the model invented.
FAQ
Can I text-to-video the carton if I describe the count carefully?
You can. You will recast the panel every take. Count, approval mark, and colourway are the expensive miss. Photograph the carton. I2V or reference-lock it so those strings are not sampled again.
Does any voice on an N95 ad kill YPP?
No. The heading is personas that present as human experts giving health advice. A narrator reading the panel, or no voice at all, is not the example YouTube wrote. A diagnosing "doctor" or a photoreal mask-wearer prescribing fit is.
If I disclose that the wearer is AI, can the channel monetize?
Disclosure does not move a video out of that bucket. Change the format. Show the carton. Keep advice off the synthetic face.
Should we I2V a clinic still and park our pack in the foreground?
No. The clinic still is a facility job. The OEM still is the carton you ship. Mixing them invents a waiting room you do not operate and a patient you should not generate. Lock the box. Overlay the count.