What kinds of video appear on a product page
On both marketplaces there are two kinds of video. A cover video plays in place of the main photo, on a loop and usually without sound. A gallery video the buyer opens themselves, once they are already looking at the product: reviews, instructions and unboxings go there.
Their jobs differ. The cover has to stop the eye in the search results within a second, while the gallery clip has to answer an objection like does it fold or is it really that small. That is why both versions are usually cut from one source.
Technical requirements of Wildberries and Ozon
Below are the references from sellers' help pages as of autumn 2026. Marketplaces change media parameters more often than photo requirements, so check your seller account before uploading a batch.
| Parameter | Wildberries | Ozon |
|---|---|---|
| File format | MP4 or MOV | MP4 or MOV |
| Gallery video | up to 3 minutes, up to 50 MB, 720-1920 px | from 8 seconds to 5 minutes, up to 100 MB, 1080-1920 px |
| Cover video | 3:4, up to 30 seconds | from 8 to 30 seconds, up to 20 MB; 1:1 or 3:4 depending on category |
| Frame rate | 25-30 | 25-30 recommended |
| What is banned | contacts, links, prices and promotions, third-party logos | the same, plus stock watermarks |
The cover-video script: eight seconds without sound
The cover is the main photo, moving. The product appears in the first frame, with no intro cards or logos, and performs one action: it unfolds, pours, rotates or goes on.
- The first frame has to work as a still photo: before the video starts playing, that is exactly what the buyer sees.
- One movement instead of a story. Eight to fifteen seconds are enough to show a mechanism or a texture.
- A loop with no jump: the last frame flows back into the first, otherwise the seam shows up as a jerk.
- Keep what matters in the centre and the upper part of the frame: the bottom of the preview carries the price and the interface buttons.
- The check is simple: watch the clip on a phone with the sound off and ask yourself whether it is clear what the product is.
Where AI saves money and where you still need to shoot
Generation handles the surroundings well: the interior around a sofa, the kitchen around a blender, the beach around a swimsuit. The product itself is better taken from real photos, otherwise the model invents seams, connectors and lettering.
The working scheme: product photography of the item from several angles, then generation from photos with references. Models with multi-reference support keep the same item consistent across scenes noticeably better than generation from a text description.
Filming is still needed where the buyer decides on tactile details: how clothes sit on a person in motion, how a cream feels on the skin, the sound and the response of a device's buttons. A generated texture that does not match the real thing turns into not like the video reviews and returns.
Which models fit product pages
For most product pages two generators are enough: one for cheap short scenes, the second for cases where the packaging has to stay readable.
| Job | Model | Why |
|---|---|---|
| rotating and showing a product from a photo | Kling 3.0 | video from a photo, clips up to 15 s, about $0.11 per second at 1080p without audio |
| a readable label or lettering | Veo 3.1 Fast | holds text in frame, about $0.12 per second at 1080p |
| one product across several scenes | Seedance 2.5 or Wan 3.0 | plenty of references, a long scene in one pass |
| cheap motion drafts | MiniMax H3 | about $0.08 per second at 768p |
What a product-page clip costs
Generation is the cheapest part of the job. A 10-second cover at 1080p on Kling 3.0 without audio is about $1.10 for a good pass; with failed attempts it comes to a few dollars.
The real money goes into the script written around buyer questions, frame selection, editing for 1:1, 3:4 and 9:16, and a moderation check. When a range has dozens of products, it pays to build a template once and ship clips in batches: the cost per clip then drops several times over.
If you want a unique clip with every frame worked through, handcrafted production starts at $190 per video up to 90 seconds.
Why a marketplace rejects a video
Most rejections come not from quality but from what is in the frame or from the file parameters.
- the captions, packaging or voiceover include a phone number, a website, a social handle or a QR code
- the video mentions price, discount or delivery terms
- generation changed the colour or the contents, and the video no longer matches the product page
- a horizontal clip padded into a square with black bars instead of being cropped
- stock footage or music without a commercial licence
- the cover file exceeds the limit: 30 seconds at 1080p with a high bitrate easily goes past 20 MB
A checklist before uploading
Eight checks that take five minutes and save a second round of moderation.
- MP4 format, H.264 codec, 25 or 30 frames per second
- cover proportions match the category, with no frames
- the product in the first frame, no intro card
- colour, model and contents match the photos on the product page
- the clip makes sense without sound, and captions are readable on a phone screen
- no contacts, links, prices or superlatives
- the cover loop has no jerk at the seam
- the file size has headroom below the limit
What next
The same source material is useful off the marketplace too: vertical videos for business for the TikTok and Reels feeds are cut from it. Batches across a whole range are the mass vertical videos format, and the full breakdown of models and prices is in the 2026 guide to AI video models.