AI product photography: what it can do and what each channel allows
What AI product photography does today, how to check an AI image against the real product, and what Google, Amazon, Etsy and seven more channels say about it.
On this page
In short
- AI image tools do four jobs for product photos: replace the background, generate a scene around the product, show clothing on a model and produce variations of an existing image.
- Google allows AI-generated product images only if the file keeps the metadata that marks it as AI-generated, the IPTC DigitalSourceType property.
- Amazon requires the keyword contains-synthetic-performer in the metadata of every image that shows photorealistic AI-generated people.
- Etsy requires one unedited photo of the actual item per listing and allows AI edits on the other photos only if the item does not look different.
- Allegro asks sellers to declare images generated or substantially modified by AI and adds an AI watermark to them; Zalando does not allow visible AI labels on images.
- Systems that resize, convert or deliver images can strip the metadata that carries the AI label, so check the final file before you upload it.
What AI product photography can do today#
AI image tools do four jobs for product images. Each starts from a real photo of your product:
| Job | You supply | You get | What can go wrong |
|---|---|---|---|
| Background replacement | A photo of the product | The same product on pure white or another plain product photo background | Edges, shadows and transparent parts change |
| Scene generation | A photo or cut-out and a description of the scene | A lifestyle image: the product in a room, outdoors or on a surface | Wrong scale, lighting or reflections; props that look as if they were included |
| On-model shots | A photo of the garment or accessory | The item worn by a generated model | Fit, drape, pattern or buttons differ from the real item; the model may need a label |
| Variations | One finished image | Other crops, sizes, angles or colors | New angles and colors are invented, not photographed |
Scene generation is the part that replaces lifestyle product photography: one studio photo becomes a trail, a kitchen or a desk without a location shoot. Channels offer some of these functions themselves. Google's image page points sellers to a tool that increases resolution, removes backgrounds and generates scenes, and eBay's listing app offers AI-generated backgrounds. So the question is not whether AI product images are accepted, but under which conditions.
The risk that matters: the image must still show the real product#
A tool that generates a scene can also redraw the product in it. Every channel with a rule on AI product images asks for the same thing: accuracy.
- Amazon: the image must "accurately represent the real scale, quantity, and color of the product".
- Etsy: all images must reflect the item's "true appearance, including color, condition, scale, and materials".
- Walmart: AI-generated images "must be truthful, accurate and not misleading".
- eBay: photos that do not accurately represent the item are not allowed.
- Zalando lists the distortions it rejects: unnatural textures or patterns, distorted text, mispositioned buttons, brand logos merged with another brand, and models or products out of scale with the setting.
Compare every AI image with a real photo, or with the product in your hand, before it goes into a listing:
- Shape. Same silhouette, same number of edges, openings and handles.
- Proportions. Length, width and height in the same relation; no stretched or thickened parts.
- Color. The color of this variant, not a nearby shade. Compare under neutral light.
- Logo and brand name. Spelled correctly, in the right place, the right size.
- Labels and printed text. Size labels, warnings, markings and care symbols legible and unchanged.
- Countable parts. Buttons, pockets, ports, straps, LEDs, eyelets: count them.
- Material and texture. Mesh stays mesh, matte stays matte.
- What is included. Nothing in the picture that the buyer does not get.
- The scene. Shadows, reflections and scale fit; the product stands on the surface, not in it.
- People. If a person in the image is generated, note it: Amazon needs a tag.
- The file. The AI label is still in the metadata (see below).
Example: a generated trail scene for the Northpeak headlamp looks right at first glance. The count shows four LEDs where the product has three, and the strap buckle is gone. The scene can stay; the product must be put back from the real photo.
What each channel says about AI images#
As of November 2026, from each channel's own page:
| Channel | What its page says | Source |
|---|---|---|
| Google Merchant Center | AI-generated images are allowed in all three image attributes. They must contain metadata that marks them as AI-generated (IPTC DigitalSourceType), and you must not remove it. Merchant Center also has a "Manage AI label" setting per image; in the EU, India, New York, California, South Korea and Vietnam, ads with assets labeled this way show a visible label | AI-generated content, AI label settings |
| Amazon | The image guide has one rule that names AI: an image with photorealistic AI-generated people needs the keyword contains-synthetic-performer in the dc:subject (XMP) field before upload. Not needed for real people, even if altered with AI, for people who are not photorealistic, or for images without people | Product image guide |
| eBay | No rule on AI-generated images on the pages read. Images must be truthful and accurate and must not infringe third-party rights; added text, artwork and watermarks are not allowed | Images policy, picture policy |
| Walmart (US) | AI-generated content must be accurate, truthful, rights-compliant and not misleading | Image guidelines |
| Etsy | One photo per listing must be unedited and show the actual item on a real background. Other photos may be edited with AI (lighting, color, crop, background) if the item does not look different. Etsy asks you to mention AI edits in the description. When it detects AI-generated images, it shows tips | Image requirements |
| Allegro | Sellers declare images that were generated or substantially modified by AI; Allegro adds an "AI" watermark to them. Allegro's developer note names 2 August 2026 as the start of the duty | Developer note (Polish) |
| Zalando | Low-quality or unnatural AI images are rejected. Visible on-image AI labels are not allowed, so content that legally needs one must not be supplied. Zalando asks for invisible marking in the metadata or as an invisible watermark, under the heading "required by December 2026" | Image guidelines |
| Microsoft Merchant Center | An optional advertiser_ai_label per offer: AIWithVisibleWatermark, AIContent or NoAIContent. It applies to the offer, not to single images, and is not available to every store yet | Product attributes |
| Kaufland | No rule on AI images on the pages read | Product data guideline |
| OTTO Market | No rule for seller images on the pages read. OTTO's terms let OTTO itself create new visualizations from your images, including with AI | Terms, section 6.14 |
Allegro and Zalando both point to the EU AI Act. Article 50 of Regulation (EU) 2024/1689 asks providers of AI systems to mark generated images in a machine-readable way, and asks those who use such systems to disclose deep fakes, which the regulation defines as generated or manipulated content that resembles existing persons, objects or places and would falsely appear authentic. In the text as published, the article applies from 2 August 2026. This page does not assess when a product image falls under it; that is a legal question.
AI tools do not change who owns the source photo. If the image you edit comes from a brand, read using manufacturer product images first.
How the AI label gets lost, and how to check it#
The label Google asks for is a metadata field inside the image file, not something visible. It is easy to lose between the tool that made the image and the URL in your feed.
What the label is. The IPTC property Digital Source Type. Google names three values to preserve: TrainedAlgorithmicMedia (created with a model trained on sampled content), CompositeSynthetic (a composite with synthetic elements) and AlgorithmicMedia. The IPTC vocabulary also has compositeWithTrainedAlgorithmicMedia for a photo edited with generative AI, for example with a generative fill. The IPTC's user guide notes that this field exists only in the XMP part of the metadata.
How it gets lost.
- Resizing and converting. Some image libraries drop metadata unless told to keep it. The documentation of one such library, sharp, says: "By default all metadata will be removed".
- Shop systems and delivery networks. The IPTC user guide warns that a site's asset management system, content management system or content delivery network "may be stripping out embedded metadata fields", often to save bytes.
- Uploads. Zalando tells partners to make sure the metadata is permanent "so it isn't stripped during upload".
How to check.
- Check the file the AI tool gave you. No label there means there is nothing to keep, and you need to add it with a metadata editor.
- Run your normal resize and export, then check the result again. The product image checker on this page reads the Digital Source Type from the file in your browser and reports whether an AI label is present.
- Download the image from the URL in your product feed and check that file too. Google fetches the image from that URL, so that is the copy that counts.
- If the label is gone, find the step that drops it and turn on its option to keep metadata.
Amazon's tag is a different field. The checker does not read it: add contains-synthetic-performer to the dc:subject field with an IPTC-compatible metadata editor, as Amazon's guide describes.
Which image type fits which slot#
| Slot | What channels expect | Where AI fits | Use a real photo for |
|---|---|---|---|
| Main image | The product alone; pure white on Amazon, Walmart, OTTO and the Allegro thumbnail; no text or props | Removing or whitening the background of a real photo | The product itself. On Etsy, keep one photo unedited |
| Lifestyle | The product in use or in a setting. Google's lifestyle_image_link asks for a real-world context, at least 600 x 600 pixels, no text | Scene generation around the real product | Anything where size or fit must be exact |
| Detail | Close-ups of material, seams, ports, labels | Cropping and light correction | The close-up itself, so that texture and stitching are the real ones |
| Scale | The product next to something of known size; eBay suggests a coin or ruler | Little | The whole image, since proportions are the point |
| Infographic | Text is restricted: none on Kaufland, Microsoft and eBay images, and OTTO requires that the information also appears in the attributes or description | Layout drafts | All text and numbers; Zalando lists distorted text among AI defects |
The size, format and background limits behind this table are in product image requirements compared. For Amazon's image slots in detail, see Amazon listing images.
Common mistakes#
- Generating the product instead of the scene. Keep the product pixels from the photo and let the tool build what is around them.
- One AI image for every variant. Google asks for an image per variant in its exact color and finish.
- Resizing after the check. Check the label on the last file in the chain, not the first.
- Treating "allowed" as "no conditions". Google, Amazon, Etsy and Allegro each attach one: metadata, a tag for generated people, an unedited photo, a declaration.
Where Central helps#
Central researches the product data a listing needs from your own data, the open web and verified databases, and backs every value with sources you can check. It generates listing images from that verified product record, and writes titles, descriptions and attributes within each channel's limits. What the sources do not support is held back.
Questions
Does Google allow AI-generated product images?
Yes, in image_link, additional_image_link and lifestyle_image_link. Google requires that every image created with generative AI contains metadata marking it as AI-generated, such as the IPTC DigitalSourceType value TrainedAlgorithmicMedia, and that you do not remove it.
Are AI product images allowed on Amazon?
Amazon's product image guide states no general rule for or against AI-generated product images. It requires every image to represent the product accurately, and it requires the metadata keyword contains-synthetic-performer on images that show photorealistic AI-generated people. Real people altered with AI do not need the tag.
Do I have to label AI product images?
It depends on the channel. Google requires metadata in the file, Amazon a tag for photorealistic AI-generated people, and Allegro a declaration, after which it adds an AI watermark. Etsy asks you to mention AI edits in the description. Zalando does not allow visible labels on the image and asks for invisible marking.
How do I check whether an image still has its AI label?
Open the final file, the one you will upload or host, in a tool that reads IPTC metadata. The product image checker on this page reads the IPTC DigitalSourceType value in your browser and tells you whether the file carries an AI label.
Can an AI product photo generator replace a photo shoot?
Not completely. Every channel with a rule requires that the image shows the real product, and Etsy requires one unedited photo per listing. AI tools fit backgrounds and scenes built around a real photo; the product itself should come from that photo.
Sources
- Image link [image_link]Google Merchant Center Helpchecked 20 November 2026
- AI-generated contentGoogle Merchant Center Helpchecked 20 November 2026
- Use AI content label settings and disclosures in Merchant CenterGoogle Merchant Center Helpchecked 20 November 2026
- Lifestyle image link [lifestyle_image_link]Google Merchant Center Helpchecked 20 November 2026
- Additional image link [additional_image_link]Google Merchant Center Helpchecked 20 November 2026
- Product image guideAmazon Seller Centralchecked 20 November 2026
- Adding pictures to your listingseBay Customer Servicechecked 20 November 2026
- Picture policyeBay Customer Servicechecked 20 November 2026
- Images, videos and text policyeBay Customer Servicechecked 20 November 2026
- Product detail page: Image guidelines and requirementsWalmart Marketplace Learnchecked 20 November 2026
- Requirements and best practices for images in your Etsy shopEtsy Helpchecked 20 November 2026
- How do Etsy's listing image requirements apply to my listings?Etsy Helpchecked 20 November 2026
- Zalando image guidelinesZalando Partner Universitychecked 20 November 2026
- Oznaczanie zdjęć AI (AI Act): aiCoCreatedContent and POST /sale/imagesAllegro Developer Portalchecked 20 November 2026
- Rules for images in the gallery and in descriptionsAllegro Help for sellerschecked 20 November 2026
- Product attributes (Microsoft Merchant Center)Microsoft Advertising Helpchecked 20 November 2026
- Product data guidelineKaufland Global Marketplacechecked 20 November 2026
- Terms and conditions for sellers on OTTO Market, version 17.0 (section 6.14, Annex 8)OTTOchecked 20 November 2026
- IPTC Photo Metadata User GuideIPTCchecked 20 November 2026
- Digital Source Type (IPTC NewsCodes)IPTCchecked 20 November 2026
- Output options: metadatasharp documentationchecked 20 November 2026
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50EUR-Lexchecked 20 November 2026