Dumpster ads are first-last: empty drive, then the bin

Two stills. Flux 3 or Veo first-last. A paragraph will not land on the second photo.

A dumpster rental ad that works is almost always the same cut: the driveway empty, then the bin on the pad. That is a bounded interpolation, not a scene you describe. If you type "show our 20-yard coming in" into a text-to-video box, the model invents a truck, a house, and a dumpster that is not yours. The second photo you already shot — the actual unit, the actual fence, the actual oil stain — never appears.

First-last frame is the job name. You hand the model where the clip starts and where it ends. The motion in the middle is generated. The endpoints are not.

A paragraph cannot land on photo B

Ordinary image-to-video knows frame one and improvises the rest. That is the wrong contract for this ad. The bin in the last frame is the product. If the generate is allowed to wander, you get a generic roll-off that will not match the photo on the landing page.

Text-to-video is worse. There is no photo A and no photo B. There is a sentence. Sentences do not pin a SKU.

The agent job is explicit: create a first-and-last-frame transition video. Attach a start photo and an end photo. It passes both as first and last frame and picks a first-last-capable row. Output is one clip that opens on the first still and lands on the second. One still is a different capability — animate a photo — and it will not honour a destination you never uploaded.

Shoot the empty drive, then the bin

Order matters because the offer is "we show up." Frame A is the empty pad: same camera height, same lens family, same time of day if you can. Frame B is the same pad with the bin placed. Do not change the house. Do not swap the weather. First-last will connect unrelated plates; the connection is the artefact everyone notices.

Keep the two frames in the same location, lighting, and lens family. A close-up of a logo as B and a wide of a street as A is not a delivery. It is a warp.

If you need a loop for Stories, use the same image in both slots. That is a different ad. The rental ad is A ≠ B on purpose.

The row is first-last, not a hero generate

Two catalog rows are built for this:

• Flux 3 first last frame to video: interpolates a start frame and an end frame, native audio, up to 1080p, 5–20 seconds. Aspect list includes 9:16. requires_image is true.
• VEO first last frame: the same two-still job on Google's row. Durations 4s / 6s / 8s. 16:9 and 9:16. Qualities through 4K. Audio on or off.

Pick the row for the length you can fill honestly. Five to eight seconds is enough for an empty drive to gain a bin. Twenty seconds of interpolation on two nearly identical stills is padding.

Do not prompt a cinematic founder talking over the driveway. That is a different ad and a different model family. The dumpster is the subject. The stills already have it.

Generate 9:16 when the destination is a story or a short. Both rows list 9:16. Cropping a 16:9 master after the fact is how the bin's edge falls off the phone.

FAQ

Can I describe the drop-off instead of shooting two photos?

No. A paragraph will not land on the second photo. First-last needs both stills. If you only have one photo, you are in image-to-video, and the model is free to invent a different bin.

Is Flux 3 or Veo the right first-last row?

Both are first-last rows. Flux 3 covers 5–20 seconds at up to 1080p with native audio. Veo first-last is 4, 6, or 8 seconds with a 4K tier. Choose the duration you actually need. Do not pick Veo 3.1 text-to-video and hope it reconstructs your unit.

What if photo B is a different angle of the same dumpster?

Then the interpolation has to invent a camera move you did not shoot. Keep A and B in the same lens family. If you want a new angle, that is a second pair, not a prompt that says "orbit."

Do I caption the clip in the same generate?

No. First-last returns the motion between stills. Burn captions after, on the file you kept. The generate is the empty drive becoming the bin. The line of copy is an editor pass.