An AI product video is worth having only if the product in it is the product you ship. Generated footage tends to drift: a label gains a letter, a bottle gets taller, a shade of green turns teal. The fix is a process that starts from your own packshots or 3D files, designs each shot as a still image first, and checks the product frame by frame before delivery.
This guide covers how that process works, which products are easy and which are hard, where the videos run and what those placements require, and the advertising rule that sits over all of it.
Why does the product change in AI-generated video?
Video models do not know your product. They know what bottles, trainers and headphones look like in general, and they rebuild the object on every frame from that general knowledge plus whatever reference they were given. Small details are the first to go. Printed text turns into plausible-looking nonsense. A logo is redrawn slightly differently at second three than at second one. Proportions stretch as the camera moves. Colour shifts with the generated lighting, which is a real problem for anything sold by shade: lipstick, paint, fabric.
On a mood film, nobody notices. On a product page, the customer is looking at exactly those details, because they are deciding whether to buy.
How to keep a product accurate in an AI product video
There is no single setting that solves this. It is a sequence of habits, and each one removes a source of error.
- Start from the brand’s own files. Packshots on a plain background, shot from several angles, are the minimum. A 3D model or CAD file is better, because it can be rendered at any angle under any light and stays exact. We also ask for brand colour references, so there is a number to check against rather than a memory.
- Design every shot as a still image first. The image-first workflow is covered in how AI reels are made. For product work it matters more than anywhere else: a still can be corrected in minutes, the product can be placed into it precisely, and the client can approve the look before any motion exists.
- Animate from the approved still. When the first frame already contains the correct product, the video model has far less room to invent. Short shots drift less than long ones, so a 20-second video is normally built from many clips of a few seconds each.
- Composite the real product where generation is not good enough. For hard cases, the scene is generated and the product is not. The real packshot or a 3D render is tracked into the footage, with generated light and reflections matched around it. The viewer sees one image. The label is pixel-for-pixel the real one.
- Check frame by frame before delivery. Someone steps through the finished video and compares the product against the reference: every word on the label, the cap, the proportions, the colour. Any frame that fails is fixed or the shot is remade. It is dull work, and it is the work that matters.
At Zero Shot, an AI video studio in London, this is the core of how we make product videos: we begin from the client’s photos, packshots or 3D files and check labels, proportions and materials before anything leaves the studio.
Which products are easy and which are hard?
| Product type | Difficulty | Why | Usual approach |
|---|---|---|---|
| Matte packaging with a simple logo (boxes, tubes, pouches) | Easy | Few fine details, predictable surface | Animate from an accurate still |
| Food and drink as a subject (pours, splashes, ingredients) | Easy to medium | Liquids and textures generate well, the pack needs care | Generated scene, pack checked or composited |
| Gadgets with clean shapes | Medium | Geometry must stay rigid, ports and buttons must not move | 3D render or still-first, short shots |
| Text-heavy labels (supplements, skincare, wine) | Hard | Small print breaks first | Composite the real label or a 3D render |
| Transparent and reflective objects (glass, chrome, perfume) | Hard | Refraction and reflections must match the scene | 3D render with matched lighting, or composite |
| Jewellery | Hard | Tiny facets, exact stone settings, customers inspect closely | Real macro photography or 3D, generated setting |
| Clothing where fit matters | Hard | Fabric drape and fit on a body are easy to fake and easy to get wrong | Real garment photography, generated background and mood |
“Hard” does not mean impossible. It means the product itself should come from a camera or a 3D file, and the AI should build everything around it. If a studio tells you a text-heavy label will be generated from scratch and stay correct, ask to see proof. That question is on our list of things to check when choosing an AI video company.
Where do AI product videos get used, and what specs do they need?
The same footage usually has to work in four or five places, so it pays to plan the versions before production rather than crop afterwards.
Shopify product pages. According to Shopify’s product media documentation, a product video can be up to 10 minutes long, up to 1 GB, up to 4K (4096 x 2160 px), in .mp4, .mov or .webm. Shopify converts what you upload and streams it at 480p, 720p or 1080p depending on the shopper’s connection. In practice a short, silent loop near the top of the gallery does most of the work.
Amazon listings. Amazon’s own guide to product videos asks for .mov or .mp4 files, up to 1080p and no larger than 5 GB. It says sellers need a Professional selling account and at least three months of selling history, that Brand Registry is not required, and that most successful videos run 30 to 90 seconds. Amazon reviews every video before it appears, and review can take several days. Length limits and content rules are set in Seller Central and do change, so read the current version there before you export.
Paid social. Meta, TikTok and YouTube Shorts placements want vertical 9:16 video, with the product visible in the first second. These are usually cut-downs with a hook added, which is a different job from the product-page loop. We cover it under AI ads for Meta, TikTok and YouTube.
Other marketplaces and retailers. Requirements vary by retailer and change often. Ask each one for its current media spec sheet and send it with the brief.
A practical delivery set for one product looks like this:
- one horizontal or square master for the product page
- one short seamless loop with no sound
- one vertical 9:16 version for social and ads
- still frames pulled from the video for the gallery
What do advertising rules say about showing a product?
The law does not care how the pixels were made. It cares whether the customer was misled.
In the UK, rule 3.1 of the CAP Code section on misleading advertising says marketing communications must not materially mislead or be likely to do so, and rule 3.11 says they must not mislead by exaggerating the capability or performance of a product. The ASA has also addressed AI directly. In its note on disclosing AI in advertising it says there is no blanket UK requirement to disclose AI use, and that a disclosure is very unlikely to fix a fundamentally misleading message. Its example is an AI image showing a cosmetic result the product does not deliver.
In the US, the FTC’s advertising guide for small businesses states that advertising must be truthful and non-deceptive, and that an ad is deceptive if it is likely to mislead reasonable consumers about something material to the decision to buy. The same guide notes that the advertising agency, as well as the advertiser, may be held responsible.
For product video this comes down to a few plain rules. Show the real size, the real colour and the real contents. Do not generate a result the product cannot produce: fuller hair, whiter teeth, a stain that vanishes. A fantasy setting is fine, because nobody believes the moisturiser lives on a cloud. A false product is not. This is general information and not legal advice, so ask a lawyer about anything borderline.
When is filming the product the better choice?
Sometimes it is. If the whole point is how a fabric moves on a body, how a mechanism feels in the hand, or how a real person reacts to using the thing, a camera will be quicker and more honest. Often the best answer is a mix: a short real shoot for the product, generated scenes around it. We set out the trade-offs in AI video or a real shoot.
If you have a product and packshots ready, send us a short brief with the places the video needs to run. We will reply with how we would keep the product accurate and a fixed price.
Questions people ask
Can AI make a video of my actual product, or only something similar?
It can show your actual product if the work starts from your own photos, packshots or 3D files and the product is checked frame by frame. A video generated from a text description alone will only ever show something similar.
Is an AI product video allowed on Amazon and Shopify?
We found nothing in Shopify's or Amazon's published video file requirements that rules out generated footage. Amazon reviews every video before it goes live, so an accurate product matters more than how the footage was made. Check the current rules in Seller Central and the Shopify Help Centre before you upload.
Which products are hardest to get right in AI video?
Products with small printed text, transparent or mirror-like surfaces, fine jewellery and clothing where fit matters. These usually need the real product composited into the generated scene rather than generated from scratch.
Do I need to tell customers the video was made with AI?
The UK has no blanket rule requiring AI disclosure in ads, but a disclosure does not rescue a misleading video. Some platforms have their own labelling rules for realistic AI content, so check each placement. This is general information, not legal advice.