Zalando AI images: no visible label, invisible marking by Dec 2026
Zalando says no on-image AI labels or watermarks, and asks for invisible marking in the file by December 2026. What that means for images made with Sume.

Zalando's rule for AI-generated product images has two parts: do not put a visible label on the image, and embed identifying data in the file's metadata or as an invisible watermark. Its image guidelines say the visible-labelling rule applies from August 2026 and the invisible marking is required by December 2026 (page updated October 1, 2026, read 2026-10-02). Sume's Image API docs do not describe embedding such marking, so check every file yourself.
This post is what the pages say, not legal advice.
What does Zalando say about labelling?
Under "Visible labelling" the page says Zalando does not allow on-image labels such as text overlays that do not show measurements, or watermarks. It also says not to provide content that requires visible labelling under applicable laws, and names deepfakes and AI-generated public-interest text as examples.
Under "Invisible marking" it strongly recommends embedding invisible marking in all AI-generated content, says you must embed identifying data within the metadata or as an invisible watermark in the file code, and says to make sure the metadata is permanent so it is not stripped during upload.
| Platform | Visible label | Marking in the file |
|---|---|---|
| Zalando | Not allowed on the image (overlays, watermarks) unless showing measurements | Metadata or invisible watermark; required by December 2026; keep it permanent |
| Google Merchant Center | Page names no visible label | IPTC DigitalSourceType, for example TrainedAlgorithmicMedia; do not remove it |
Does a Sume image carry that marking?
The Sume docs I read (Image API) cover size, format, quality, references and billing. They do not state that a returned image has embedded provenance metadata or a watermark. So do not assume it. Download a generated file, inspect its metadata with a tool you trust, and keep the result with the job id.
If the file has no marking, you would have to add it in your own pipeline. Zalando's page does not name a standard, while Google's page names the IPTC property, so a metadata-based approach is the one with a named tag today.
How do I avoid stripping it?
Zalando warns about metadata being stripped on upload, and Google says not to remove embedded IPTC tags from AI-generated images. Any resize, crop or re-encode step in your own pipeline can drop metadata, so add the marking last, after you produce the final 1:1.44 JPEG, and test by re-reading the file you will actually upload.
Also keep text off the image. A price banner or a badge such as "AI" is a visible overlay, which Zalando does not accept.
What should I do next?
Here is the short version, in the order to do it.
- Make the image with no overlays, labels or watermarks.
- Inspect a sample output for metadata before you promise compliance.
- Add or verify the marking as the final step, then upload that exact file.
- Re-read Zalando's page before December 2026; the wording may change.
Sources
Related posts
More in Use cases
- Zalando apparel model views: same model, shoes on, sleeves shown
Zalando's apparel guide for model views: same model and styling across views, shoes on, full-length sleeves, simple styling. What to put in an AI prompt.
- Zalando image rejection 117 and 123: too few views, a mannequin
Zalando rejects articles for missing images (reason 117) and visible mannequins in mandatory views (reason 123). Duplicates do not count. What to regenerate.
- Zalando image size: ask for 2000x2880, an exact 1:1.44
Zalando wants 1:1.44 JPEG portraits, best practice 1801x2600. Sume image_size takes multiples of 16, so 2000x2880 is an exact 1:1.44. Request it with jpeg.
- Zalando jewellery packshot: lower third, symmetry, back view
Zalando's jewellery guide: primary packshot in the lower third of the grid, a back view showing fasteners, no upside-down logos. A prompt checklist for Sume.
Written by Sume