The storefront wanted a video slot and all I had were still product photos

The marketplace opened a video slot on the storefront and we had no footage. Reshooting meant scheduling, so I tried turning the existing product photos into something that moves.

A few things came out of that round.

Decide which kind of motion you need

Turning a product photo into video covers three different jobs that people talk about as one.

Type	What moves	Length	Suits
Camera move	The camera pushes, pulls or pans; the product stays still	3 to 5 seconds	Storefront slots, covers
Product motion	The product itself moves, fabric drapes and so on	3 to 8 seconds	Apparel and soft materials
Multi-image sequence	Several stills cut together with transitions	8 to 15 seconds	Multiple angles or colourways

For a storefront slot the camera move is the safest. The product does not move, so it cannot deform, which is the failure mode buyers notice first.

Cost and length go together

TangyuanAI's pricing page lists Seedance 2.0 at $0.040 per second (pricing page, checked 6 August 2026).

At that rate a five second clip is about $0.20. That sounds trivial until you multiply by the retry factor. Producing the storefront clip, we averaged three to four attempts before one was usable, which puts the real cost per usable clip between $0.60 and $0.80.

That is still far below a reshoot.

The input image matters more than expected

Turning a still into video means the model has to infer what sits behind the parts it cannot see. So the input quality decides the output.

The subject must be complete. If the product is cropped at the frame edge, whatever the camera reveals on that side is invented.

Resolution has to be sufficient. Upscaling a low-resolution still for a camera move looks obviously soft.

Background should be clean. Cluttered backgrounds distort during the move. A plain background is the most stable.

Three checks on the output

Look at these in order and reshoot if any fails.

First, check the product for deformation. This is the least acceptable failure, and it shows most on products with straight structural lines, where edges bend during the move.

Second, look for things that were not in the original. When the model fills occluded regions it invents content.

Third, look at the first frame. A storefront slot shows the first frame when nothing is playing, so a poor first frame makes the rest irrelevant.

Frame size

Storefront slots are usually 16:9 or 9:16 depending on the marketplace. The product image tool supports both, along with 1:1, 4:5 and 3:4 (product image page, same date). Generate at the target ratio rather than cropping afterwards; cropping damages the composition.

Limits

The video must not demonstrate a function the product does not have. Camera moves and lighting are presentation. Functionality is not.

Results depend on the uploaded material and the brief. There is no guaranteed success rate.

Sources

Model pricing from the TangyuanAI pricing and credits page, https://tangyuanai.vip/en/pricing . Capability and frame sizes from https://tangyuanai.vip/en/about and https://tangyuanai.vip/en/main-image . All checked 6 August 2026. Retry factors and the check list come from my own notes.