Zalando AI distortions: a checklist to run before upload

Zalando's image guidelines name three groups of unacceptable AI distortions: an inaccurate representation of the article, an implausible setting, and an inconsistent model appearance. The page says Zalando reserves the right to reject any AI-generated asset that breaks believability, shows irregularities or fails to represent the product accurately. The list is in the image guidelines, updated October 1, 2026 (read 2026-10-02), and turns directly into a review step for generated images.

What are the three distortion groups?

Each group has examples on the page. Use them as the review criteria rather than judging by feel.

A practical way to run this is a two-person pass on a sample: one person checks the garment or product details, the other checks the setting and the model. Keep the failing images and the reason beside the job id, because the same failure tends to repeat with the same prompt, and the notes tell you which instruction to change. Zalando itself retouches positioning, crops and background colour but says it will not make edits that change how the article looks, so it will not fix a distortion for you.

How do I turn that into a review step?

Put the source photo and the output side by side and check, in order: print or pattern placement, buttons and zips, logo spelling and shape, stitching, then shadows and reflections. Then look across the whole set of views for one article: does the product look identical in each? The page lists inconsistency across the set as a distortion, so one good image is not enough.

Write what must not change into the prompt, for example keep print, buttons, logo and stitching exactly as in the photo, but treat that as a request, not a guarantee. Product exact in AI video and retouching cover the same habit.

What about the model views?

If you generate a person, the page says to keep facial features, body proportions and skin tones consistent across images of the same model, and to avoid airbrushed skin. For apparel, the apparel guide says to use the same model and styling for all views of an article. A fresh generation per view makes that hard, so reuse one reference image across the set via input_references, and reject any view where the face or build drifts.

What should I do next?

Here is the short version, in the order to do it.

• Compare each output with the real product, not just with the prompt.
• Review the whole image set for one article together.
• Drop any view with warped text, logos or buttons, and regenerate it.
• Keep the source photo and job id so a rejection can be traced.