AED cabinets: wall still plus overlay on this wall, no fake emergency
Photograph the cabinet on the real wall, overlay the location, and do not generate a rescue.
Photograph the cabinet on the real wall. Overlay location. Do not generate a rescue.
An AED cabinet is a facility plate. The buyer already knows the box: the alarm strip, the glass or polycarbonate door, the wall it hangs on, the paint colour next to the lift. The job is to show this install, then typeset the floor and the wing. A generated collapse in a lobby is a patient you did not consent, a clinician you did not hire, and a health video YouTube already named. Shoot the wall. Burn the line. Leave the emergency off the generate.
This wall is the plate
The honest still is a photograph of the cabinet where it actually hangs. Same corridor, same mounting height, same adjacent door frame. The facilities team will recognise the tile. A paragraph will invent a nicer lobby.
Do not send “photoreal AED cabinet in a hospital hallway, ready for use” to a still model and call it the install. Imagen 4 is Google’s text-to-image row: 4K, 4 credits, photoreal, text-only. It cannot ingest the wall photo. It will decide a wall. That wall is not yours. Use Imagen 4 only when you need a closed, empty SKU still of the cabinet you sell — door shut, no body, no responder — and you are willing to say the picture is a product render, not the site.
Home security installers own nearby wall hardware: cameras on a real property, coverage, a cable run. An AED cabinet is not that page. It is a medical device on a public wall. Dentists own the clinician side: office tour, a named team. This post is the facility plate. Do not rewrite either /for. Do not borrow a talking hygienist to “explain” a cabinet.
If the only photo includes a person, crop them or shoot again. A generated patient is a hard stop. No residual limb, collapsed extra, or dummy in scrubs. The cabinet is the subject. The wall is the address.
Overlay the location you already posted
The location is a string the building already uses: “Level 2, east lift lobby.” “Staff kitchen, left of the fire panel.” Write that string. Do not ask a model to letter it into the glass.
Add timed text overlays to a video burns lines you wrote at start_sec and end_sec. Nothing is transcribed. Floor, then wing, then “do not block” is three authored beats on a locked plate. Font, size, colour, and position apply to the set, not per line. If you only have one persistent badge for the whole clip, that is add a text overlay: one string, whole duration, top / center / bottom.
Auto-captions listen. A mute wall still has nothing to transcribe. Running a transcript pass hoping “AED” appears is a spend on silence. Prompted type inside Imagen 4 is a misspelled floor you cannot edit.
Crop to the ratio you will ship before you overlay. Top / center / bottom sit in a different safe zone after a 9:16 reframe. Picture lock, then type. Offers and floor maps change. The cabinet does not.
Imagen 4 is a cabinet still, not a rescue
Imagen 4’s job, when it is in this pipeline at all, is a still of the box: closed door, green sign, empty interior if the door is open in the brief. Four credits. 4K ceiling. Text-only input. That is a SKU render. It is not a training film.
Do not prompt:
- A person on the floor
- Hands opening the cabinet onto a dummy
- A “first-aid instructor” walking the steps
- A child, a wheelchair, a hospital bed next to the box
Those are people. Health content that invents a body is not a catalog still. If you need motion on the real wall — a slow push, a light rake, no new corridor — photograph the install and animate that file on an image-to-video row. The still is the contract. Motion is allowed. A rescue is not.
A 4K invented lobby is still the wrong wall.
A fake emergency is a health persona
Synthetic presenters on health and finance are a YPP kill. Help’s third inauthentic-content bucket is AI-generated personas that present as human experts on health. A constructed responder telling a stranger how to pad, shock, or “you should” is that shape. Labelling the face AI does not leave the bucket. Fan funding and ads sit behind Partner Program eligibility.
A silent cabinet on a real wall is not that shape. Authored type on the plate is typesetting, not a mouth. A real, named, credentialed trainer on camera is a different day and a different file — and then you caption that take. Generation can still do the box. It cannot be the consult.
Third-person description of the plate is allowed: the cabinet, the height, the floor overlay. Prescription from a synthetic face is not. If the first three seconds look like a rescue, rewrite until they look like a wall. Ship the install. Overlay the location.
FAQ
Can we generate a CPR demo so people know how to use the cabinet?
No. That demo is a patient and a protocol. Photograph the closed cabinet on the real wall. Overlay the location and the training URL you already host. A real instructor, named, on a different shoot if you need a voice. Do not invent a body so the cabinet has a scene.
Why not Imagen 4 the wall if we do not have a photo yet?
Imagen 4 is text-only. It cannot hold this corridor. It will invent tile, paint, and a cabinet that is not the serial on the work order. Get the photo. Use Imagen 4 only for a closed SKU still you will label as a product render, never as the site.
One overlay or timed lines?
One persistent location badge is a single overlay on the whole file. Floor, then wing, then a “do not block” beat is timed overlays, with timestamps you set. Neither job transcribes. Neither job belongs in the Imagen prompt.
Is a talking AI first-aid instructor acceptable if we disclose it?
Disclosure does not move a constructed health expert out of the YouTube bucket. Keep the facility plate. Keep the overlay. If someone must speak a protocol, that someone is a real named person, not a generate.