Doorbells: porch still; no generated neighbors
I2V the device on a real door you own.
I2V the device on a real door you own. Do not generate the street. Do not generate the neighbor who "just happens to walk past." A porch ad is a product shot with a house attached. The house has to be yours or a set you control. The people on the sidewalk are private figures, not extras.
A deepfake of a private person is not a style. "We changed the face 15%" is not a defence. TikTok's private-figure rule is permission, not a label. Background is not a likeness carve-out for a recognisable person.
The porch still has to be a door you own
Mount the unit. Photograph it. Daylight, the actual trim, the actual bell, the actual app screenshot if the app is in the ad. That frame is the SKU. A generated Victorian with a generated brass button is a different product, and a practitioner (or a homeowner) will notice the hardware that is not yours.
Lock the still before you spend on motion. If the plate is the wrong colourway, the wrong wedge, or a door you do not have rights to, fix the still. Video credits are a terrible identity debugger.
Do not prompt "a cute family opening the door in golden hour" and hope the device survives. Text-to-video will re-decide the hardware and invent a household. You do not have a likeness grant from that household. You also do not have a pack shot.
If you need a second angle — three-quarter on the button, night IR, the chime inside — shoot it on the same door. Then you have two stills, not a prompt.
Neighbors are not extras
A generated pedestrian, a generated kid on a bike, a generated "package thief": private people, or fictional people presented as traffic on a real street. Leave them out.
If a demo needs a person — ringing, dropping a box — use someone you employ or hired, on a door you control. Most doorbell brands do not need that. A crew hand you filmed is enough. Minors are a hard stop. Do not generate a child at the door. Package-drop clips can be the device, the mat, and the box.
Do not scrape Street View or a listing photo of someone else's porch and I2V it as a set. That is their house.
Animate the device, not the street
Seedance 2.5 Image to Video is the still-to-motion row: one approved porch frame, continuous clip, four seconds to thirty, up to 720p, native audio with the picture. A second image can set the last frame when the move has a destination — a push that must land on the button, a night still that must hold the IR look.
Image-to-video is the tool shape. Prompt only the motion: a slow push on the unit, the LED breathing. Do not prompt a passer-by. Seedance I2V is still the still-to-motion job because the SKU has to remain the SKU.
App UI belongs as a real screenshot, overlaid or I2V'd on its own still. If you need a line ("Person at door," a CTA), burn a text overlay after picture lock. Do not ask the model to invent a notification. The street can stay empty. Empty is how you avoid a neighbor.
FAQ
Can we generate a package thief to show the alert?
No. That is a generated private person committing a crime in your ad. Use a box on a mat you own, the real alert UI, and a caption you wrote. Film a crew member if you need a hand.
Why not text-to-video a perfect porch?
Because T2V re-decides the device and invents a house. Hardware, LED, and trim are the expensive failure. The porch still is the cheap place to lock them. I2V is how you stop paying for a second doorbell.
Does the clip need people to feel like a home?
No. The unit, the door you own, and the app still are the product. People who are not on payroll are a likeness problem. A crew hand you filmed is optional, not required.
Can we I2V a customer's install photo?
Only with a grant that covers that photo and any generation from it, and still without neighbors in the frame. Prefer a door you own. Customer porches come with customer families.