Car dealership video ads with AI, from your lot photos

Car dealership video ads can be made without filming every car: animate the photos you already took of each car on the lot into short clips, with the real photo as the first frame, then add a voice-over with the car's model, price and offer and export a vertical and a wide cut. Generated motion can change badges, wheels and trim, so check every frame against the real car before the ad runs.

The Sume steps come from the Video generation, Timeline 1.0, Video inspect and Video captions docs and the Sume API reference, read on 2026-09-29.

What should a dealership video ad show?

One car per ad, shown as it sits on your lot. A short inventory spot is three to five shots:

• The hero angle: a front three-quarter photo with a slow push-in.
• Details from your own close-ups: the wheel, the interior, the dash.
• The offer: model, trim, price and what's included, said in the voice-over.
• Only cars you have. Don't generate a car, a color or a trim that isn't on the lot, and don't let the prompt add options the car doesn't have.
• Offer and finance wording is your call: this post doesn't cover the disclosure rules where you advertise.

How do I turn a lot photo into a video clip?

Send each photo at a public HTTPS URL as the first_frame of a POST /v1/videos job and prompt one camera move, such as a slow push-in; images sent as input_references only guide the model. AI car video generator walks through that request. For an ad, generate at aspect_ratio 9:16 when the vertical cut matters most, send resolution explicitly, and read each model's options from GET /v1/videos/models.

How do I check that the car still looks right?

Run POST /v1/video-inspect on each finished clip. With frames omitted it returns 8 stills spread across the clip, and the probe and stills are unbilled. Compare each still with the photo: badge, grille, wheel design, trim and any lettering. If one drifts, regenerate with a smaller move or cut the shot. Product logo warping in image-to-video covers frames versus references in more depth.

How do I add the price and offer?

Make the voice-over with Sume text-to-speech, then join clips and voice in a Timeline 1.0 render. Every URL in the render must already be a media.sume.com file in your workspace; completed Sume jobs, such as the clips above, return their outputs there.

• Set audio.duration_seconds to the voice-over's length; it is the output length.
• Render twice from the same slots: output.width × output.height 1080 × 1920 for Reels and Shorts, then 1920 × 1080 for a website or YouTube, with fit: "blur" on vertical clips in the wide cut.
• In current code the render keeps only the voice spine and an optional soundtrack; each clip's own sound is dropped.
• For the price on screen, burn authored cues (text, start, end) with POST /v1/video-captions. Today the caption job refuses a video over 60 seconds or with no audio stream.

How much does a dealership video ad cost?

You pay per step from your workspace USD balance. For one car: