AI Labeling

All results you generate with Tevyon come from AI models. They show what the models can do, not real people, products or events. On our home page, these are the results under “Made with Tevyon”, each with its model. The other photos and videos there are real footage from the royalty-free archive Pexels. They show areas of use, not results from Tevyon.

For every result, we store which model created it and when. That way you can prove its origin later.

Several model makers embed machine-readable labels in their results: invisible watermarks such as SynthID in Google models, or provenance data under the open C2PA standard (“Content Credentials”). Tevyon leaves these labels in the file: we change neither picture nor sound, and we store a file with Content Credentials byte for byte as the model delivers it, so that its signature remains valid.

In addition, Tevyon writes a machine-readable label into every new image and video: metadata under the XMP standard with the IPTC source type “trainedAlgorithmicMedia” (generated by a trained AI model) and Tevyon as the tool. The image itself does not change. Files with Content Credentials don’t get this label, because their maker’s label already identifies them as AI.

You can check whether a file contains Content Credentials with freely available C2PA verification tools. If you edit or export a file with other programs, such data can be lost. A label is therefore an indication, not proof.

We are working on also giving all results Tevyon’s own Content Credentials. As soon as that is in place, it will say so here.

Art. 50(4) of the EU regulation on artificial intelligence (AI Act) requires everyone who publishes a “deepfake” to disclose that the content has been artificially generated or manipulated. A deepfake is an image, video or audio that resembles real people, objects, places or events and can falsely appear to be authentic. The obligation applies if you distribute the content in the EU, including from Switzerland.

• Label such content clearly, e.g. with “AI-generated” or “made with AI”, in the image, in the caption or in the post.
• For obviously artistic, satirical or fictional works, a notice that does not get in the way of enjoying the work is enough, for example in the credits.
• Use the AI labeling of the platforms you publish on, if they offer one.
• Don’t remove watermarks or provenance data to conceal the origin.

Under Swiss law, too, missing labeling can be unlawful, for example as misleading advertising under Art. 3 of the Swiss Unfair Competition Act (UCA) or as an infringement of personality rights under Art. 28 of the Swiss Civil Code (CC). The rules on real people are in the Usage Policy.

If you talk to an AI assistant in the Studio or through a connector, it is recognizable as such. If an assistant writes texts that you publish to inform the public on matters of public interest, you must label them as AI-generated, unless a person has reviewed them editorially and takes responsibility for them (Art. 50(4) AI Act).

If you have found content that comes from Tevyon and is deceptive, report it to us. How to do that is explained under Reporting content.