Garage epoxy: the car in the shot has to be the customer's car
I2V from their photo. A Kling sports car is a different garage.
I2V from their photo. A Kling sports car is a different garage. The floor you poured sits under a specific vehicle on a specific slab. If the clip parks a generated coupe where their daily driver lives, you did not advertise the job. You advertised a showroom that is not on their street.
The still the customer already sent — the car in the bay, the oil stain, the water heater in the corner — is the plate. Motion is allowed. A prettier car is not.
The driveway photo is the car
An epoxy quote is a site. The buyer knows the bumper, the ding in the door, the rake of the apron. Text-to-video does not. Ask for "a sports car in a glossy garage at golden hour" and the model will pick from every cinematic bay it has seen. Average is a Kling sports car. Average is not their Civic, their van, or their truck.
Seedance 2.5 Image to Video is the still-to-motion row: one still in, a continuous clip out, four to thirty seconds, native audio with the picture, 720p as the ceiling on that endpoint, optional last frame when the move has a destination. Frame one is the file they sent. Labels on the epoxy chip chart, the hose bib, the VIN plate in the windshield — they survive because they were already in the pixels.
Prompt only camera and physics. A slow push across the flake. Light sliding on a floor that is already that colour. Do not re-describe "a midnight-blue supercar in a luxury garage." That paragraph restages the job. Image-to-video starts from the upload. A sentence does not.
Lock the still before you spend on motion applies twice: the floor has to match the pour, and the car has to match the owner. If the phone snap is muddy, shoot again. Do not prompt the car out and hope the model rebuilds an honest bumper.
A Kling sports car is a different garage
Kling O3 Pro text-to-video does not require an image. It is reasoning-enhanced generation with camera control and native audio, three to fifteen seconds. That is how you get a cinematic coupe pulling into a bay that never existed. Useful for a brand film about "the idea of a garage." Fatal for a jobsite ad.
The failure mode is a paragraph about a supercar because Kling is good at cars. Identity of the lot is not a camera-control problem. It is a missing plate. Image-to-video versus text-to-video is the same fork. The cost of being wrong is a clip that shows someone else's vehicle on their floor. Homeowners notice cars.
Register the vehicle as a product asset
If the week has more than one scene — before the grind, after the flake, a pull-out to the street — upload the car once. Set up reusable characters and products opens an Asset Builder card for a product (or a location). Uploading reference images does not spend generation credits. The asset is reusable by key across every scene in the workflow. You do not re-upload the bumper per shot.
A full cast at once is the usual start: the car, the finished floor still, the crew van if it is actually in frame and you are allowed to show it. Then generate. The stills are the spec. Language does not pin a VIN.
A last frame is control when you have two approved photographs of the same bay — empty slab, then the car on the cured floor, same standing spot. It is not a prompt for "a nicer car." Unrelated plates will connect, and the connection is the artefact.
Motion the floor, not the VIN
Once the customer photo is locked, the generate may:
- Drift the camera along the flake
- Add light that does not rewrite geometry
- Land on a second approved still of the same bay
It may not restyle the bumper, grow a second storey, swap the daily driver for a catalogue coupe, or invent a lift the shop does not have. Those are claims. The owner will clock them before they clock your logo.
If you do not have their photo yet, get one. A phone still in the open door is enough. Then animate that. Do not migrate off I2V because a Kling coupe looks expensive this week. The expensive coupe is a different garage.
FAQ
Can I prompt Kling to "match this customer's car" from a sentence?
No. A sentence is not a plate. Upload their photo to image-to-video and describe only the camera. Kling text-to-video will still invent a sports car.
What if the only photo includes a neighbour's car on the street?
Keep it if it is actually in the shot and you are not selling their bay. Crop if the quote is only this slab. Do not generate a cleaner version of a private vehicle you did not photograph.
Do I need a last frame for a before-and-after pour?
Only when both stills are theirs — the same bay, the same standing spot, empty then finished, or dirty then coated. A single hero still with a slow move is the usual job. A last frame is control, not a default.
Should I onboard the car as a workflow asset if I only have one clip?
Not required. One still into Seedance I2V is the whole job. Onboard the car when you will reuse it across scenes so you are not re-uploading per generate.