AI content

AI product photography for ecommerce: what to check

You get what AI does well with product photos, the image rules for Google Shopping, Meta ads and the EU, and the checks every AI image needs before it goes live.

AI product photography works best when you start from a real photo of the product. AI then places it in new scenes, adds lifestyle context and makes a version for each channel and ad format. Check every image against the real product: shape, label, colour, size and text. Keep the AI metadata for Google Merchant Center, and label realistic AI images in the EU.

Key takeaways

  • What AI does well: new scenes, lifestyle context and format variants, all made from one real product photo.
  • Accuracy: check shape, label, colour, size and packaging text against the real product in every image.
  • Google: Merchant Center requires the IPTC tag trainedAlgorithmicMedia in every AI-generated product image.
  • Main image: Google does not allow prices, calls to action, watermarks, borders or added logos on it.
  • Labels: from 2 August 2026, the EU AI Act asks you to disclose AI images that could pass as real.
  • Bottom line: start from the buyer and the angle, make 3 to 5 variants per angle, and let a person approve every image.

AI product photography uses for online stores

AI product photography is best at the work around a real product photo. It places the product in new backgrounds and scenes, shows it in use and makes a version for each channel and ad format. It also makes short product videos from photos. It is weakest at the product itself, so small text, exact colours and fine details need a check every time.

The input decides the result. AI can draw a product that looks like yours, but it will not be your product. Give it clean photos of the real item from several sides, in daylight, with the label easy to read. The better the photo, the less the AI has to invent.

  • New backgrounds and scenes: your product on a kitchen table, a desk or a shelf, made from one studio photo.
  • Lifestyle context: the product in use, so the buyer sees its size and its purpose.
  • Format variants: the same image in 1:1, 4:5 and 9:16, one for each channel.
  • Short product videos: a few seconds of movement made from still photos.
  • One style for the whole catalogue: the same light and the same colours on every image.

Product image rules by channel: Google Shopping, Meta ads and the EU

Each channel has its own image rules, and an AI image must pass them like any other image. Google checks the main product image and the AI metadata in the file. Meta sets the size and the safe area for each ad format, and it labels some AI ads. In the EU, two laws cover realistic AI images and misleading product images.

The table lists the rules we check before an image goes live. Rules change, so each row links to its official page in the sources at the end of this article.

Image rules to check by channel
Channel Image type Rule to check
Google Merchant CenterMain product image (image_link)Shows the exact item you sell, at least 500 x 500 pixels. No prices, calls to action, watermarks, borders or added logos.
Google Merchant CenterEvery AI-generated image (main, additional and lifestyle)The file keeps the IPTC DigitalSourceType tag trainedAlgorithmicMedia. Do not remove it.
Facebook Feed adsImage adMeta’s ads guide lists a 4:5 ratio at 1440 x 1800 pixels.
Instagram Reels adsVertical video or image9:16 at 1440 x 2560 pixels. Keep text and logos out of the top 14%, the bottom 35% and 6% on each side.
Meta adsAds made or heavily edited with Meta’s AI toolsMeta adds an “AI info” label. It also detects some ads made with other AI tools.
EU AI Act, Article 50Realistic AI images and videos that could pass as realFrom 2 August 2026, the company that publishes them must disclose that they are AI.
EU consumer law (Directive 2005/29/EC)Every product image in an ad or on a product pageThe image must not mislead the buyer about the main features of the product.

Product accuracy checks for AI product images

Every AI image of a product needs one check. It must show the product the buyer will receive. Compare it with the real item or a real photo, side by side and at full size. An AI model draws the product again in each image, so a detail that was right in one image can be wrong in the next one.

This is also a legal point. Under the EU Unfair Commercial Practices Directive, a practice is misleading when its overall presentation deceives the average consumer about the main features of a product. The directive names features such as composition, quantity and specification. A wrong colour or an extra part in an ad image can do exactly that.

  • Shape and size: the product has the same form and proportions, and its size fits the scene around it.
  • Colour: the colour matches the real item and the colour name on the product page.
  • Label and packaging text: read every word and every number, letter by letter.
  • Parts and quantity: no extra buttons, straps or handles, and the same number of items in the pack.
  • Materials: wood looks like wood, metal looks like metal, and the finish is the same.
  • Logo: your logo sits where it sits on the real product, and nowhere else.

Google Merchant Center rules for AI-generated product images

Google Merchant Center has one extra rule for AI-generated product images: the file must keep its AI metadata. Google’s help page says all AI-generated images must contain the IPTC DigitalSourceType tag trainedAlgorithmicMedia. The rule covers the main image, the additional images and the lifestyle image. You must not remove the tag from the file.

The main image has stricter rules. It must show the exact item you sell, with a clean view of the product. Google does not allow prices, calls to action, shipping details or promotional words on it. Watermarks, borders, barcodes, added store logos and products that are not part of the sale are not allowed either. The image must be at least 500 x 500 pixels.

Our rule is simple. The main image stays a real photo, or a clean cut-out of one. AI scenes and lifestyle images go in the additional and lifestyle image fields. Before you upload, open the final file and check that your export and compression steps kept the metadata.

Clean images and complete product data are also the base of our ecommerce SEO and AI search work, where we prepare product pages and feeds for Google and AI search.

AI labels for product images and ads: Meta and the EU AI Act

Two kinds of AI labels matter for an online store. Meta adds its own “AI info” label to some ads. The EU AI Act asks for disclosure of realistic AI images and videos from 2 August 2026. Neither rule bans AI images in ads. Both make sure that people can know when AI made an image.

Meta adds the “AI info” label to ads made or significantly edited with its own generative AI tools. It also detects some ads made with other AI tools through industry-standard signals. When Meta’s tools make no significant edit and add no photorealistic person, Meta adds no label. Since an update in June 2026, the label appears in the “About this ad” panel.

Article 50 of the EU AI Act splits the work in two. The companies that provide AI image tools must mark the output in a machine-readable format. Tools already on the market before 2 August 2026 have until 2 December 2026. The companies that use the tools, such as a store, must disclose a deep fake. The AI Act defines it as AI content that resembles real persons, objects, places or events and would falsely appear authentic.

The law does not name product photos. A realistic AI photo of a real product is closer to that definition than a clear illustration. For evidently creative work, the duty is lighter: you flag the AI content in a way that does not spoil the work. Our practice is to label AI content clearly in every case, and to ask a lawyer when a case is unclear.

Angle-first method for AI product ads

Start every batch of AI ad images with the buyer and the angle, not with the tool. An angle is one reason to buy, said in one way. Write 3 to 5 angles, make 3 to 5 variants of each, and test which angle sells. AI makes variants fast, so the limit is the thinking before it.

Our angle-first method

Updated November 2026aimo

Choose the buyer and the angle before AI makes one image.

Five steps for every batch of AI ad images and videos, from the buyer to the test.

  1. Buyer Name the person who decides and the problem they want to solve.
  2. Angles Write 3 to 5 reasons to buy. Each ad carries one angle.
  3. Variants AI makes 3 to 5 images or videos per angle from your real photos.
  4. Check A person compares every variant with the real product and approves it.
  5. Test Run the angles side by side and keep the angle that sells.

This is our own method at AI for Marketing Ops. The ranges are working rules, not measured results.

Sources: AI for Marketing Ops content method, November 2026

Variants that change only a colour or a word test nothing. Each angle should change the argument. Here is an example for a reusable water bottle, with three angles a store could test against each other.

Every variant of an angle shows the same true product. The scene, the person and the text change. The product does not.

This is how our AI content creation service works. We define the buyer and the angles first. Then AI makes the variants, and a person on our team reviews each one before you approve it.

  • Money angle: one bottle instead of the plastic bottles you buy every week.
  • Gift angle: a bottle with a name on it, as a present for a colleague.
  • Everyday angle: the same bottle at the gym, at the office and in the car.

AI photos and videos for RNJ Transport

For RNJ Transport, a road freight forwarder in Klaipėda, Lithuania, we made the website photos and the social media videos with AI. It is a logistics company, not a store, but the method is the same: one style, real facts and a person who checks every piece before it goes live.

The 27 website photos share one daylight documentary style, made with AI image models. They show no logos and no faces, so no image can be taken for a real customer or employee. The director’s portrait is a real photo. We did not generate a real person.

For LinkedIn, we made 20 videos, 10 in English and 10 in Lithuanian, at 1080 x 1350 pixels and 18 to 26 seconds each. That is the same 4:5 shape as a Facebook Feed ad. For YouTube, we made promo films and Shorts.

The RNJ Transport case study shows the full project. The photos and videos were part of a website and brand revamp with a new multilingual site.

Work that stays with people in AI product content

AI makes the images, but people decide what goes live. A person approves every image, checks every claim in the ad text and handles anything that involves a real person. We never give these jobs to an AI model, for our clients or for our own work.

If your team wants to make and check this content itself, our AI training for commercial teams runs on your own products and your own ads.

  • Approval: someone who knows the real product approves every image and video before it goes live.
  • Claims: every number, size and promise in the ad text is true, and you can prove it.
  • Real people: get written consent, a model release, from anyone you photograph.
  • AI people: an AI model or presenter never pretends to be a real customer, and never copies a real person.
  • Rights: no copies of other brands’ ads, photos or products.

Common questions

Can I use AI-generated images in Google Shopping?

Yes, if the file keeps its AI metadata. Google Merchant Center requires the IPTC DigitalSourceType tag trainedAlgorithmicMedia in every AI-generated image, in the main, additional and lifestyle image fields. The main image must also show the exact item you sell, with no prices, promotional text, watermarks or borders.

Do I have to label AI product photos?

On Meta, the platform adds an “AI info” label to some ads itself. In the EU, from 2 August 2026, the AI Act asks you to disclose AI images and videos that could pass as real. In Google Merchant Center, the label is the metadata in the file. We label AI content clearly in every case.

Can AI make a product photo without a photo of the product?

AI can draw a product that looks like yours, but it invents the details. The buyer then receives something different from the image. Start from real photos of the item from several sides, with the label easy to read. Then let AI build the scene around it.

Can AI make product videos from photos?

Yes. AI can make short videos for ads and social media from product photos. Check the whole video, because the product must look the same in every frame. For RNJ Transport, we made 20 LinkedIn videos of 18 to 26 seconds each, in English and Lithuanian.

How many ad variants should I make with AI?

We start with 3 to 5 angles and 3 to 5 variants per angle. Each angle is a different reason to buy. A few clearly different angles teach you more than many variants of one idea. This is our own working method, not a rule from an ad platform.

Can I show my products on AI-generated models?

You can, with care. Label the images as AI where a buyer could take them for real photos. Never present an AI person as a real customer, and never make one look like a real person. For clothing, check the fit and the length against the real garment.

Sources and checks (9)
  1. Google Merchant Center Help: image link [image_link] (checked November 2026)
  2. Google Merchant Center Help: preserving metadata tags for AI-generated images (checked November 2026)
  3. Meta Ads Guide: Facebook Feed image ads (checked November 2026)
  4. Meta Ads Guide: Instagram Reels ads (checked November 2026)
  5. Meta: Expanding GenAI transparency for Meta’s ads products (February 2025, updated June 2026)
  6. European Commission, AI Act Service Desk: Article 50, transparency obligations
  7. European Commission, AI Act Service Desk: Article 3, definitions (point 60, deep fake)
  8. European Commission, AI Act Service Desk: implementation timeline (checked November 2026)
  9. Directive 2005/29/EC on unfair commercial practices, Article 6

Book a 30‑minute discovery call

Tell us what you sell and to whom. We check your business in Google and AI search before we talk. If we are not a fit, we tell you.

Book a discovery call

30 minutes · Free · Online