Zalando bans non-English text on primary views: check AI images

Zalando's image guidelines list non-English text on primary views as a restriction (read 2026-10-02). Image models can add lettering you did not ask for, or garble a label, so run a text check on every generated primary packshot: nothing readable beyond the product's own English print, and nothing else added.

The restriction list

These are the key restrictions on Zalando's image guidelines (read 2026-10-02).

Where AI images go wrong

Garments often carry printed words: a size tag, a care label, a logo. When a model redraws them, letters can change or a word in another language can appear. The Image API docs advise escalating quality for finals, dense text and packaging (Image API), which helps legibility but does not remove the review step.

The page says text on primary views. Whether a brand's own printed logo counts is a question for your Zalando contact; the page as read does not say.

A text check per image

For a batch, make this a manual gate after the jobs complete, since a Sume job reaching completed says the image was made, not that it is correct (Jobs and results).

• Zoom to 100% on the packshot and read every printed or woven word.
• Compare each word with the real garment; any word that is not on it is a failure.
• Look for stray captions, price tags and watermarks.
• Reject and regenerate rather than paint over; keep the job id of the version you ship.

Making the check scale

For a large batch, run OCR over each primary packshot and flag any image that returns text, then review only the flagged ones by eye. OCR will miss some and misread others, so treat it as a filter that cuts the pile, not as the approval. Keep the original photo beside the generated one so the reviewer can compare in a second.

Related rules

For Zalando's AI labelling rules, see the labelling post. A mannequin in a generated shot hits rejection reason 123.