AI Transparency
Last updated: 7 May 2026
Nexvy is an AI content-generation platform. Everything you create here — images, video, audio, music, and chat responses — is produced by artificial intelligence. This page explains what we disclose, what we mark, and what we cannot mark, so you can judge for yourself how far our signals go.
It also explains how this relates to Article 50 of Regulation (EU) 2024/1689 (the "EU AI Act"), whose transparency obligations apply from 2 August 2026. Nexvy acts as a provider of an AI system under that Article. If you publish what you create here, you may separately act as a deployer, with obligations of your own — see Section 5.
1. You are always interacting with an AI
Nexvy Chat and its assistants are AI systems, not people. The assistants have job-title names (for example "Scriptwriter" or "SMM Manager") and illustrated avatars, but there is never a human on the other side.
Our assistants are instructed never to claim to be human, and to answer truthfully and immediately if you ask whether you are speaking with a human or an AI. Every chat screen carries a persistent notice to the same effect. The same applies in our Telegram bot and in the Telegram Mini App.
AI output can be inaccurate, incomplete, or biased. Review it before you rely on it or publish it.
2. How we mark AI-generated content
Images generated on Nexvy carry an embedded, machine-readable marker recording that they were produced by a generative AI model. We write it as an XMP metadata packet using the IPTC standard field DigitalSourceType with the value "trainedAlgorithmicMedia" — the same interoperable vocabulary used across the industry — alongside the model used, the creation timestamp, and the generation identifier.
The marker contains no personal data. We deliberately do not embed your prompt, your name, your email address, or your account identifier, because the file travels to third parties every time it is downloaded or shared.
We also preserve any provenance metadata already embedded by the upstream model provider rather than stripping it, so signals such as Content Credentials or provider watermarks survive our processing wherever the provider supplies them.
Files downloaded from Nexvy are additionally named "nexvy-ai-generated-…" so the origin is visible without opening the file.
3. What we cannot mark, and why
We would rather tell you the limits than imply a guarantee we cannot keep.
- Video, audio, and music outputs are not yet marked. Those formats need container-level writers that our image tooling cannot provide; the work is in progress and scheduled to complete before 2 December 2026.
- Chat text is not marked. There is no machine-readable marker for plain text that survives copying and pasting, and no industry standard proposes one. Instead, the chat interface itself is persistently and clearly labelled as AI (see Section 1).
- Metadata is fragile. Re-encoding, screenshotting, cropping, converting formats, and uploading to most social platforms will strip our marker. Some platforms strip all metadata on upload as a matter of routine.
Because of this, the absence of a Nexvy marker is not evidence that a file was made by a human. It usually means the file was processed somewhere along the way.
4. Visible watermarks
Images generated on the Free plan carry a visible "nexvy.ai" watermark. This is a plan feature, not a legal disclosure, and it is separate from the machine-readable marking described above — the machine-readable marking is applied on every plan, paid or free.
5. Your obligations when you publish
If you publish or distribute AI-generated content, the law may require you to disclose that fact — in particular for deepfakes (content that resembles real people, places, or events), and for AI-generated text published to inform the public on matters of public interest. Under Article 50(4) of the EU AI Act that duty falls on you as the deployer, not on us.
Our Acceptable Use Policy requires you to make that disclosure where the law demands it, and prohibits presenting AI-generated content as authentic human-created media in a deceptive or unlawful way. Deceptive deepfakes of real people are prohibited outright.
Do not remove or obscure the markers we embed.
6. How to check a file
Our marker lives in the file's XMP metadata and can be read with any standard metadata tool, including ExifTool, by inspecting the Iptc4xmpExt:DigitalSourceType field.
For content generated elsewhere, or where our marker did not survive, the model provider may have applied its own signal — for example Google's SynthID or C2PA Content Credentials. We do not control upstream watermarking and cannot verify or vouch for content produced outside Nexvy.
No detector, ours included, can reliably prove that a piece of content was made by a human.
7. Contact
Questions about this page or about how we mark content: [email protected].
This page describes our current practice and is updated as our marking coverage expands.