AI Transparency Notice

1. The photographs are AI-generated

Your personal model is trained on the photographs of your face and then produces new images: your likeness in clothing, a setting, a pose and a light that you selected. None of those scenes ever took place.

Because the result is photorealistic and depicts a real, identifiable person, it falls within what Regulation (EU) 2024/1689 (the AI Act) calls a deepfake. We say this plainly rather than hiding it: it does not make the images illegitimate, it makes them subject to transparency rules.

2. How we mark them

Machine-readable marking, in every file: before a generated photograph is ever stored or shown to you, we write an XMP metadata packet into the image file. It carries the IPTC property DigitalSourceType with the value trainedAlgorithmicMedia — the standard code for content created using generative AI — together with the creator tool, a credit line naming Portrevo and the date of generation. This is exactly what Article 50(2) of the AI Act asks for: marking in a machine-readable format, detectable as artificially generated.

It is genuinely detectable. Anyone can read the marking with a standard metadata tool such as ExifTool or the IPTC Photo Metadata Viewer, without asking us for anything. Some platforms read this property and may add their own AI label when you upload the image; whether they do, and whether they keep the metadata at all, is entirely their decision and can change at any time.

Visible watermark: yes, on the unpaid previews. Before you pay, every generated photograph is shown to you in the application reduced to 600 pixels on the longer side and covered by a visible Portrevo watermark. Once you have paid the one-off unlock, the same photograph is delivered to you clean — no visible watermark, no visible AI label, only the machine-readable marking described above, which is written into the stored file and is therefore present in every file you download. The visible watermark is not a measure under the AI Act: Article 50(2) requires machine-readable marking only, and the European Commission's guidance on Article 50 of 20 July 2026 states expressly that the marking obligation is limited to machine-readable formats and that perceptible labels are a complementary measure, not a requirement for us. The watermark protects our business model, in the same way as any photo library protects its previews; we do not burn it into the product you have paid for.

Please do not remove it. You agree in section 15 of the Terms of Service not to strip, alter or obscure the marking we embed, and section 12 of the Terms separately prohibits removing, cropping out or retouching away the visible watermark on an unpaid preview. Be aware that most social networks delete all image metadata when you upload a file — that is their processing, not ours, and it is one of the reasons your own disclosure duty in section 4 below matters.

What is not in the file yet: the metadata is not digitally signed (C2PA / Content Credentials) and there is no invisible watermark in the pixels. Both are planned as a second layer of marking. The watermarked preview is a reduced copy re-encoded on our server for each request; it is not a delivered file and it does not carry the metadata packet. The legal requirement of Article 50(2) is met by the metadata marking of the delivered files described above; the two additions go beyond it.

3. What you must not do with them

Do not use the photographs to deceive anyone about their nature. In particular, do not present them as authentic photographic evidence, do not use them where an authentic photograph is legally required (identity documents, visa or licence applications, insurance or court evidence), and do not use them to suggest that an event, a place or a meeting took place.

Do not use them to impersonate another person or to create the impression that someone endorsed something they did not.

The full list of prohibited uses is in section 12 of the Terms of Service.

4. Your own transparency obligation when you publish

The two duties are split. Article 50(2) of the AI Act binds us as the provider of the AI system, and we discharge it by marking the file machine-readably, as described in section 2. Article 50(4) binds the deployer — the person who actually publishes the content. When a Portrevo photograph gets published, that person is you, not us.

In plain language: if you use the photograph in a professional, business or commercial context — a LinkedIn profile, a CV or job application, a company website, a freelance portfolio, a conference speaker page, business correspondence — you should say next to it that it was generated by AI. One sentence is enough, for example: profile photo generated by AI (Portrevo). This applies from 2 August 2026.

If you use the photograph purely privately and non-professionally — a dating app, a personal account you do not monetise — the AI Act does not place this duty on you at all (Article 2(10)).

You cannot discharge that duty with our metadata. The Commission's guidance is explicit that a deployer cannot rely on the machine-readable marking embedded by the provider: the disclosure has to be perceivable by a human being, clearly and distinguishably, at first exposure at the latest. Metadata is not. That is why the sentence has to come from you.

In the application we note this in a short line under the gallery, with a link to this page. We cannot give you legal advice about your particular publication.

5. Interacting with an AI system

The whole Portrevo pipeline is automated. Photo quality checks, prompt assembly, training, generation and safety filtering are done by software; no person selects or retouches your images as part of the normal service.

We do not operate a chatbot or a virtual assistant. If we ever add one, it will identify itself as an AI system from the first interaction.

6. Questions

Questions about how we mark AI content, or a request for information for your own compliance documentation: kontakt@portrevo.com.