Solar screens: window still with the screen SKU

I2V. The solar-screen clip that has to sell is a photograph of a real window with the real screen in it — mesh, colourway, the SKU the quote already named — moving just enough to prove the product sits in a frame. It is not a generated house. It is not a generated mesh.

Photograph the install. Animate the photograph.

The window still is the SKU

Solar screens are judged on the opening they cover. Buyers zoom for weave, edge, and how much of the yard remains visible. Text-to-video will invent a prettier porch with a melted spline. That is a failed SKU, not a curb-appeal upgrade.

Image-to-video keeps frame one as the file you sent. Seedance 2.5 is the family for that still-to-motion row: a locked plate, four to thirty seconds in one pass, native audio, 720p on the image-to-video endpoint, optional last frame when the shot has to land on a second approved photo — exterior with the screen in, interior looking out through the same SKU. Prompt only the camera. A slow push toward the mesh, a cloud that was already in the sky. Do not prompt "a Mediterranean villa at golden hour with insect screens."

If the still is the install photo from the job, use that photo. If the still is a studio shot of the roll against a window buck you own, use that. Do not generate a second house, a second paint colour, or a second weave. A generated screen with the wrong geometry is a quote problem sitting in a Reel.

I2V from the installed screen, not a generated house

The product-video path exists for the same reason. AI product video runs reference-to-video: two or three clean angles of the actual screen — roll, spline, the labelled box — while the scene is allowed to move. Use it when the SKU has to appear in a setting you do not have a photo of. Do not feed a finished install hero to reference-to-video and ask it to "just add motion." It will restage the house.

I2V is the contract when the composition is the ad: this window, this mesh, this light. Reference-to-video is the contract when the object has to travel. Pick one. The split, in one sentence each, is image-to-video vs reference.

Captions and SKU strings belong in a text overlay you wrote. Video models do not typeset "80% UV" onto a spline. Burn the spec after picture lock.

Last frame is a second approved photo

A before/after that is two stills you already have — bare opening, then the same opening with the screen — is a first-last job, not a paragraph about transformation. Both frames have to be photographs of the same opening, or the interpolation will invent a third house.

Lock the still before you buy seconds. If the install photo would not ship as a quote attachment, motion will not save it. If it would, describe only movement. Identity is already in the frame.

Keep the neighbour's house, the street number, and the family in the kitchen out of the generate. Those are people and addresses. The SKU is the screen.

FAQ

Can I text-to-video a "typical suburban window" and drop the SKU on later?

You can, and you will get a second window. Composite never quite matches the spline the model invented. Start from the install still. I2V is the row.

Does Seedance 2.5 image-to-video keep the weave readable?

It holds the photograph you sent as frame one. Fine mesh is still the first thing motion smears. Keep the screen large in frame, keep the shot short, and check the take at full size. If the weave fails, shorten the move — do not prompt "sharper mesh."

What goes in the last-frame slot?

A second photograph you already approved: interior looking out, dusk on the same elevation, the roll on the buck. Not a prompt for "the backyard at night."

Is a generated installer in the shot a good trust cue?

No. An invented person on a ladder is a likeness and a testimonial you did not collect. Show the screen. If you need a face, overlay a tech you actually recorded.