N52 Join
Product Studio Contact
Sign in Try it free Zu Deutsch wechseln

AI disclosure

This translation is provided for convenience. The German version is the legally binding one.

Every image produced by the N52 platform (n52.ai) is created with a generative AI model. Under Article 50(2) of the EU AI Act, we mark each of these images in machine-readable form. This page describes which markings an image carries and how you can verify them yourself.

Three markings in every image

Content Credentials (C2PA). Every image we deliver carries a digitally signed and timestamped manifest based on the open C2PA standard, identifying it as AI-generated (digitalSourceType: trainedAlgorithmicMedia). For images delivered unchanged, this is the model provider's manifest (OpenAI). If the platform reworks an image after generation, n52.ai signs a manifest of its own that includes the model provider's manifest as proof of origin. Any later change to the image data invalidates the signature and is therefore detectable.

IPTC metadata (XMP). The marking is additionally present as plain-text metadata in the image: Iptc4xmpExt:DigitalSourceType with the IPTC value trainedAlgorithmicMedia. This field is read by common image applications, digital asset management systems and platforms such as Google Merchant Center.

Invisible watermark (SynthID). The image model embeds a watermark into the pixels that is not visible to humans. It survives even when metadata is stripped — for instance through cropping, re-compression or screenshots.

How to verify an image

  • Content Credentials: Upload the image to contentcredentials.org/verify. The page shows the creator, the time, the AI-generated marking and the chain of provenance. Alternatively, the free command-line tool c2patool checks any file locally; the specification is openly available at c2pa.org.
  • IPTC metadata: With ExifTool: exiftool -XMP-iptcExt:DigitalSourceType image.png. The meaning of the value is documented in the IPTC vocabulary.
  • Watermark: OpenAI's verification tool detects the SynthID watermark and the Content Credentials of images created with OpenAI models.

What the marking is not

The marking is not visible lettering in the image, and it does not restrict how you use your images. It is the machine-readable evidence required by law that an image was created with AI. Anyone publishing images that depict real people, places or events in a deceptively realistic way must additionally disclose this visibly under Article 50(4) of the AI Act.

Questions

If you have questions about the marking, you can reach us at support@n52.ai.