Product video is the part of a listing buyers skip past and then notice. A grid of stills flattens every item to the same size and the same light; a moving clip gives the eye somewhere to go, which is why marketplaces keep pushing short video above the fold and why a good one keeps earning attention long after the thumbnail has been seen. The good news is that the production cost has collapsed. You no longer need a turntable, a studio or a model with steady hands. You need one clean still, an upscale, one controlled camera motion, and a frame-by-frame check that the label still reads the way it did before the render touched it.
That last step is where AI product video quietly fails, and it fails in a way you only catch at full size. Generative models do not read type, they reproduce the shape of type. A can of coffee looks perfect on a wide shot until the render reaches the fine print, where the ingredients block dissolves into glyphs that look plausible at thumbnail scale and become nonsense on a product page. Logos drift. Small legal text becomes a texture. If the clip is going to sit next to a buy button, where a buyer can zoom, someone has to look at every frame that contains text and compare it against the original still.
The four steps, in order
The pipeline is short, and the order is not arbitrary: each step makes the next one cheaper. Shoot or source one clean still. Upscale it while it is still a still. Animate a motion small enough for the label to survive. Verify every frame, then export one cut per marketplace. Skipping the upscale means paying for it twice, because moving footage is far harder to sharpen than a static frame, and a soft source gives the model more room to invent detail exactly where your text should be. Skipping the verification means the clip goes live with a corrupted claim on it, which costs more than every render in this article combined.
| Product category | Recommended motion | Model |
|---|---|---|
| Bottles and cans | Slow thirty-degree orbit, static camera, no product wobble | Kling |
| Apparel on a hanger | Gentle sway of the fabric only, no body movement | Wan 2.5 image-to-video |
| Boxed goods | Slow push in toward the front panel | Veo 3.1 |
| Cosmetics with fine print | Light sweep across the surface, product locked in place | Seedance |
| Furniture | Slow pull back to reveal scale in the room | Veo 3.1 |
| Jewellery | Static shot, one glint travelling across the metal | Kling |
Upscale before you animate
The crystal upscaler starts at 40 credits per run, priced by output size, and returns a larger, cleaner source. Run it on the still, never on the clip, so the model is not guessing at glyph edges it can barely see.
Animate one motion you can defend
A product animation prompt is one sentence of movement and one sentence of camera, and almost nothing else. Appearance words are wasted: the still already carries the product, the light, the lens and the style, so every adjective you add competes with the pixels and drags the render toward a colour shift. Say what the product does and how much of it there is. Say what the camera does, even when the answer is nothing at all. Then add an explicit instruction to preserve the label, because it costs one clause and occasionally it works. Wording like subtle, slight and barely nudges motion down; spins, whips and explodes push it up. High motion is where type breaks apart and where plastic surfaces start to look liquid.
Animate the still
Upload the upscaled product shot, paste a one-action prompt, and see the credit cost before the run. Video models run 60 to 600 credits depending on which one you pick, and provider-side failures are refunded in credits.
Start from an example
Verify every frame, then export per marketplace
Play the clip at full size rather than in a preview window, and step through it frame by frame. Every frame where the label is legible has to match the original still character for character. When it does not, the fix is almost always less motion and a tighter crop, never a longer prompt, because more words give the model more chances to redraw what you wanted kept. Then cut one version per surface: square or four-by-five for the gallery, nine-by-sixteen for the feed. The vertical crop is the one that eats text, so re-check after cropping rather than trusting the master. Keep the model output as the master, export the derivatives from it, and name files by SKU so a re-run six weeks from now cannot quietly overwrite an approved clip.
Frequently asked questions
- What is the fastest way to make an AI product video for an online store?
- Shoot or source one clean still with the product filling most of the frame, upscale it before animating, then run a single controlled motion such as a slow orbit or a light sweep. Verify the label on every frame that shows text, and export one crop per marketplace from the same master.
- Why does the text on my product label break in AI video?
- Generative models reproduce the shape of type rather than reading it, so fine print, ingredient lists and small logos get redrawn during the render. The practical answer is to keep motion low, add an instruction to preserve the label, and inspect the frames that contain text at full size.
- Should I upscale before or after animating?
- Before. A static frame sharpens cleanly and gives the model a clearer source, while moving footage is much harder to repair. The crystal upscaler costs from 40 credits per run, priced by output size, and should be the last thing you do to the still before it becomes a clip.
- How much does one product clip cost?
- Video runs cost 60 to 600 credits depending on the model: Sora 2 is 60, Kling 90, Veo 3.1 is 300, Wan 2.5 from 300, Seedance is 500, and Wan Animate and Wan 2.2 Animate are 600. The studio shows the exact cost before each run, and failures on the provider side are refunded in credits.
- How long should an ecommerce product clip be?
- As short as the motion allows. Clips are typically five to ten seconds depending on the model, which is enough for one clean orbit or one light sweep. Write one action that completes inside the clip instead of a sequence that gets cut off mid-move.
Turn a still into a clip
Open the studio, upload an image, describe the motion in one line and pick a video model. The credit cost is shown before the job runs.