Wheelchair OEM: the chassis still is the catalog, not a generated rider
Seedance identity on the frame photo. Do not invent a rider.
Seedance identity on the frame photo. Do not invent a rider.
A wheelchair listing is a chassis: tube geometry, caster trail, cross-brace, upholstery, the badge on the side frame. That photograph is the catalog. A generated person in the seat is not a lifestyle upgrade. It is a private figure you did not shoot, often a residual-limb invention, and on health-adjacent pages it reads as a patient. Photograph the frame. Animate it. Leave people you do not have a grant for out of the chair.
The chassis still is the listing
Factory SKU stills already exist. Someone photographed this configuration for the line sheet, the 510(k) packet, the dealer PDF. That file is the product. If the footrest hangers, the spoke count, or the serial plate is wrong, every second of video is a mis-ship.
Image-to-video starts from the file you upload. Frame one is the catalog or lab still. Prompt only camera and physics — a slow orbit that keeps the cross-brace readable, a hold on the caster, light moving across chrome that is already in the picture. Do not prompt "a smiling rider rolling through a park at golden hour." That restages the SKU as a campaign still you do not have, and it invents a person in a medical device.
Text-to-video of "a lightweight wheelchair turning on seamless white" re-decides the frame. A prettier cousin with melted welds is a failed SKU, not a cinematic upgrade. Use the manufacturer pack shot of the actual configuration. Do not generate a second colourway, a missing anti-tip, or an imprint the real row does not carry.
Seedance holds the frame; it does not cast
Seedance 2.5 is the still-to-motion family: four to thirty seconds, native audio, 720p, optional last frame when the shot must land on a second approved angle of the same configuration. Image-to-video is the job: one approved still in, a continuous clip out. Identity of the chassis is why you are here.
A last frame is control, not a default — folded then open, or three-quarter then the serial plate. Do not invent a lifestyle landing with a person in the seat the factory never shot. Thirty seconds in one pass is how a badge does not drift across a stitch of short text-to-video clips. Need a five-second dealer GIF? Generate five seconds. Crop the warehouse aisle before you buy seconds. Native audio rides along — write room tone or replace the bed. Do not send a talking presenter to this row. On a wheelchair that mouth is a health-adjacent persona.
Nearby /for pages are a different buyer
Physical therapists is a clinician surface: recovery arcs, form checks, clinic tours. That page is not an OEM catalog. A PT practice that I2Vs your chassis jpeg and captions it as last week's seating eval has used the wrong still. The clinic's proof is the room they photographed and the named human who agreed to appear. An OEM who copies the PT playbook — generated patient, invented gait, a white-coat host prescribing a cushion — has used the wrong still the other way.
The manufacturers /for page is plant-floor scale. Useful if you are filming your jig with a shot list engineering already cleared. It is not a substitute for the isolated chassis still a dealer needs. Do not rewrite either /for into this brief. OEM needs the piece, isolated, matching the card, and no invented wearer. A hand in frame is a likeness without a grant. A generated rider is a person in a medical device you never hired.
Overlay the SKU. Do not generate a clinician.
Model names, seat widths, weight limits, and HCPCS-shaped strings belong in a text overlay you wrote. Video models do not typeset a mobility SKU. They invent a plausible plate. Pause the clip and read the badge. If the letters are the model's, you shipped a cousin.
A photoreal talking presenter on a wheelchair is a health-adjacent YPP pattern. Synthetic presenters on health and finance are a YPP kill: AI personas styled as human experts giving advice. A constructed clinician walking seating, pressure, or "what you should order" is that bucket. Labelling the face AI does not leave it. Facility and SKU stills plus authored type are the stack. No generated patients. No residual limbs. No fake clinicians.
Health claims stay on the approved labelling. The clip can show the object. It cannot promise that a generated rider's mobility improved.
Minors are a hard stop. Do not generate a child in a chair. Do not "age up" a school photo. Do not put a generated kid in the background of a family hallway.
FAQ
Can I generate a generic rider if I never used a real photo?
A fully invented face is still a bad idea on this listing: it reads as a patient and as a testimonial. Film a real person with a replica grant, or show the chassis.
Should OEM use the physical-therapist /for page as the brief?
No. That page is recovery content for a clinic. OEM is I2V of the approved frame. Link the clinician page when you are the clinic. Do not copy its patient-progress templates onto a line sheet.
Why not text-to-video a "lifestyle roll" and composite the SKU later?
Because text-to-video will invent a second chair in the shot, or a rider, or both. Start from the chassis still. If you need a room, shoot or generate an empty plate and overlay type. Do not ask a video model to invent the person the still was meant to avoid.
Is 720p enough for a dealer orbit?
For feed, a product-page loop, and most buyer WhatsApp clips, yes. The frame matching the spec matters more than a higher-resolution invention of a different chair. Do not switch to text-to-video to chase resolution.