eBay stills are the auction photo; do not regenerate the item
I2V from the listing. A new T2V item is a different SKU to the buyer.
I2V from the listing. A new T2V item is a different SKU to the buyer. The auction photo is the contract. Motion is allowed. A second object is not.
A marketplace buyer matches a picture to a parcel. Generate a new object from a text prompt and you have made a different SKU: new seams, new colour, new label, new wear. That file can look better and still be a lie about the item you will ship. Keep the photo already on the listing. Animate it.
The listing still is the contract
Treat the auction photo the way a factory treats a drawing. Crop it. Light it. Shoot a second angle of the same object if you need one. Do not ask a text-to-video model to “make a cleaner version of this jacket.” That sentence is a new listing.
The failure mode is not ugly footage. It is a beautiful shot of something you do not have. Buyers compare pixels to the item in their hands. A regenerated twin fails that comparison even when the prompt named the brand and the colourway. If the listing photo is bad, reshoot or recrop the real item. Then animate that file.
Animate the photo you already posted
AI image to video is the job: the upload is frame one, not a style hint. Labels, wear, and layout survive because they are already in the file. Prompt motion — a slow orbit, a light shift, a hold — not a new product. Listing stills are already a named use case on that tool.
Seedance 2.5 covers the image-to-video row: a single still into a native clip up to 30 seconds at 720p, optional last frame, native audio if you want it. That is a long enough listing loop. It is not a reason to throw the listing photo away.
- Prompt motion, not the object. “Slow orbit left, table static” is a brief. “Premium leather sneaker on marble” is a new listing.
- Do not first-frame a prettier twin. If you generate a still and then animate that, you have already left the auction photo.
- Last-frame is a camera move, not a redesign. A second image is for where the shot lands.
Length is a dial. Pick the seconds the placement needs. Do not pick a new object because the prompt box is empty.
A text-to-video item is a different listing
Veo 3.1 is the right row for a short cinematic scene with native audio and a 4K tier — 4 / 6 / 8 seconds. That is a brand film. It is not an auction photo. Running Veo “in the style of my listing” still samples a new object. Buyers do not buy style. They buy the thing in the still.
For most listings you do not need a mountain. You need the same photo, moving. When the take is done you still have a clip, not a posted listing. Crop and export after. Do not ask the generate to typeset the SKU title into the pixels.
FAQ
Can I clean the background and still call it the listing photo?
Only if the object is the same file, edited, not regenerated. A crop that leaves the item untouched is still the SKU. A text-to-video “studio shot” of a similar object is a different SKU.
Why not generate a hero still, then I2V that?
The buyer compares to the listing, not your hero. I2V from a new still is video of an item you do not have. Reshoot the real item. Then animate that.
Is Seedance 2.5 required for listing video?
No. It is one I2V family with a 30-second pass and an optional last frame. The rule is the input: the listing still as frame one. Do not pick a text-to-video row because it looks more expensive.
Do I need native audio on an auction clip?
Usually not. A silent orbit of the real item is the listing. Do not generate a voiceover that claims features the still does not show.