AI Disclosure
Last updated: 26 September 2026
Hespera is a creative studio built on artificial intelligence. This page explains where AI is used, how you and the people you write to can tell, and what remains your responsibility.
It is published under Article 50 of Regulation (EU) 2024/1689 (the AI Act), which has applied since 2 August 2026.
1. Where AI is used
Text, images, video, speech and design suggestions in Hespera are produced by AI models. So are the answers given by Vesper, the assistant inside your account, and by the assistant that replies to your customers in chat where you enable it.
The models come from several providers. Which one is used depends on the function and on the model you select. The current list of providers is published at hespera.ai/subprocessors.
2. When you are talking to an AI
2.1. Vesper identifies itself as an AI assistant in the interface where it is used.
2.2. Where the assistant answers your customers — in the chat widget on your storefront, or through a connected messaging channel — every conversation carries a disclosure that the answer comes from an AI assistant. It is shown on the first reply and repeated at least every twenty-four hours, in the language of the reply.
2.3. This disclosure cannot be switched off. You may choose whether it appears at the start or the end of the message. The obligation is on us as the provider of the system, not on you as its user, and a system that could be configured to hide it would not meet the requirement.
3. How AI-generated files are marked
3.1. Every image, video and audio file created in Hespera carries machine-readable content credentials recording that it was generated by AI. The marking follows the C2PA specification and uses the IPTC value for media produced by a trained algorithm.
3.2. The marking is applied to everything leaving the Service through its media pipeline, not to a list of particular models. This is deliberate: a list would fall behind the models, and a file that slipped past it would be unmarked.
3.3. Files that you upload yourself are not marked as AI-generated, because they are not.
3.4. Content credentials can be removed by anyone who edits a file in other software. That is a property of the format, not a defect of ours, and it is why the marking is a signal of provenance rather than a guarantee.
4. What we do not do
4.1. We do not add invisible watermarks to pixels. The law does not require it, and the techniques available would degrade files by orders of magnitude.
4.2. We do not mark generated text machine-readably. There is no interoperable way to do so that survives copying; the law requires marking only as far as it is technically feasible.
4.3. We do not permit model providers to train on your content.
5. Your responsibility when you publish
5.1. Where you publish content that depicts a real person, place or event in a way that appears authentic but is not — including anything made with face or body try-on — the law requires you, as the person publishing it, to disclose that it is artificially generated or manipulated.
5.2. That obligation is yours and cannot be discharged by us. The content credentials we attach are not a substitute for telling your audience.
6. Limits of AI output
AI output can be wrong, outdated or unsuitable, and can resemble existing work. Review anything before you publish it or send it to a customer. You remain responsible for what you publish, as set out in the Terms of Service.
7. Contact
Questions about this page: support@hespera.ai