The roof before/after is Seedance, not a founder talking head

A solar install is a still of the house then the house with panels. Image-to-video holds the property. A Veo talking founder is a different ad.

The buyer is not shopping a personality. They are shopping this roof. If the house drifts, the ad is a lie even when the founder is charming. Route the before/after as still-to-motion. Save speech for a cut that is actually speech.

Two stills of the same house

Start with a photograph of the property you already installed, or the survey still you would show in a proposal. That file is frame one. A second still of the same geometry with panels on it is the destination, not a prompt you hope a text-to-video model invents.

Image-to-video keeps the upload as the opening frame. The motion prompt describes camera and weather — a slow drift, late light on the array — not a paragraph that re-describes the house. Seedance 2.5 Image to Video is the catalog row built for that contract: one still into a continuous clip at up to 720p, durations from four seconds to thirty, native audio on the take. You can add a second image as the closing frame when the shot must land on the panelled roof.

The published family page is Seedance 2.5. The I2V row is a variant on that family, not a talking-head model. Thirty seconds in one pass is why you do not stitch four short Veo clips around a roof and pray the gutter line still matches.

720p is the ceiling on this endpoint. Feed work lives there. A 4K founder plate is a different row.

The founder line is not the roof

Veo 3.1 is the dialogue job: native audio, 4 / 6 / 8 second clips, 720p / 1080p / 4K. That is a talking founder, a warranty line, a "we pull the permit" read. It is a good ad when the mouth is the shot. It is a bad way to prove the array sits on that house.

Mixing the jobs is how solar ads go wrong: Veo invents a plausible house that is not the customer's; a talking-head model is asked to carry the offer over a roof still; a founder take is treated as proof of workmanship.

Proof of workmanship is the still you already shot on site. Motion may add sun and camera. It may not rebuild the property. If you need both, split the slate: hold the roof on Seedance I2V, cut to Veo for the line, stitch. One generate that tries to be both will either invent a house or mute the founder.

Hold the property, then decide if anyone talks

The test is brutal and short.

1. Would you print the still on a yard sign? If no, stop. Video will not repair a bad survey photo.
2. Does the shot need to remain that house for longer than a GIF? If yes, I2V, not text-to-video.
3. Does anyone have to speak a claim? If yes, that beat is Veo (or a voiceover you attach later). It is not a motion prompt on the roof.

Do not write "our founder walks the roof explaining the array" into an image-to-video prompt. That is a presenter brief wearing a property still. Native audio on Seedance I2V is a mix on the roof take — wind, a bed. Write silence or write the cue. Mouths are not this job.

FAQ

Can I just generate the house with panels from a text prompt?

You can. You will get a house. You will not get the house. Labels, roof pitch, neighbour trees, and the street the buyer recognises are the expensive failure. Lock the still. Then animate it.

Is Veo 3.1 wrong for solar ads?

No. It is wrong as the roof. Use it when the ad is a spoken line at 4 / 6 / 8 seconds, especially if you need the 4K tier. Use Seedance I2V when identity of the property is the product.

Do I need a last frame as well as the empty roof?

Only when the destination must match a second approved still — empty roof to panelled roof, wide to array close-up. A single hero still with a slow camera move is the usual job. A last frame is control, not a default.

Why not run the founder talking over the roof in one model?

Because that is two jobs. Identity lock and always-on dialogue live on different rows in this catalog. Stitching is cheaper than asking one generate to hold the gutter line and the mouth.