authustle
2 min read

Product Video for Small Online Stores: What a Clip Costs and Which Model to Use

You have product photos and no budget for a shoot. The three formats that convert, the cheapest models that produce them, and the honesty line you cannot cross in commerce.

A small store has product photos, a catalogue that changes, and no appetite for a shoot per SKU. Generated product video is one of the few AI applications where the economics are not arguable — there is no face to protect, so the cheap models are genuinely good enough.

The three formats worth making

1. The hover. The product floats, rotates slowly, light shifts across it. No transformation, no story. It is the workhorse: a scroll-stopper built from an asset you already have.

2. The transformation. Squish, melt, inflate — deliberately unreal, obviously stylised, nobody is deceived. Works because the texture problem that plagues photoreal generation simply does not apply.

3. The context shot. The product in a room, on a table, in use — a setting you never photographed.

The cheap models are the right models here

No human identity to preserve means the premium tiers buy you nothing. Prices checked against the live endpoints on 3 August 2026:

Thirty-five to fifty cents per product clip. For a fifty-SKU catalogue that is under $25 to give every product a video, which is less than the shipping on one return.

Start at Kling 2.6 Pro or LTX 2.3. Move up only if a specific product needs it.

Prompting products

Describe the sequence, not the label. "The bottle sits still on the marble counter, then rotates slowly as the light moves across the glass" beats "floating product effect", because the model has no idea what your app calls the effect.

Two levers do most of the work: name the material (glass, matte plastic, brushed steel) and name the light (window light from the left, hard studio key). Ambiguity in either produces the flat plastic look that marks a low-effort clip.

Five seconds. One beat.

The line you cannot cross

A stylised transformation is obviously not real and deceives nobody. A generated shot that misrepresents what the product is or does is a different thing entirely — it is a claim a customer will test when the parcel arrives, and it is the kind of ad creative platforms check.

Concretely: do not generate a colour you do not stock, a size relationship that is false, a texture the product does not have, or a result it does not produce. The refund and the chargeback cost more than the shoot you avoided.

Meta requires disclosure of AI-generated content in ads, and enforcement can reach retroactively into a campaign already running. Budget for the label rather than for hiding it.

Where to spend if you have more

The step up is not a better video model — it is a better input image. A clean product photograph on a plain background, correctly lit, produces a better clip on a seven-cent model than a phone snap does on a fourteen-cent one. Fix the photo first.

ecommerceproduct videoshopifyltxbudget

Keep reading