Legal
AI Transparency Notice
Last updated 13 November 2026
1. Who this notice is for
This notice tells you where you deal with AI on ewpire, operated by LeMans Labs OÜ (registry code 16872044, Valukoja 8/1, 11415 Tallinn, Estonia), and how AI-generated output is marked. ewpire is the provider of the AI system it offers under its own name; the models inside it are developed by other companies, which are named in the model catalog. If you publish or distribute that output, you have duties of your own, set out in Section 4.
2. You are interacting with AI (Art. 50(1))
Every mode of ewpire – Chat, Code, Image, Video, Music, Consensus and, where ewpire makes it available, AI Visibility – is powered by AI systems, not human operators, and the product says so next to every input field. Answers in Chat and Consensus and every generated image, video and track show the model that produced it; Code answers do not name the model yet.
Choosing a model. You can pick a model yourself or let Auto pick one. Auto selects a model for the task; outside Code, the model it used is shown with the result. A model you choose yourself is shown before the request runs. If that model is unavailable, Auto may use another. A model you choose yourself is not replaced, except that a Consensus synthesis may be written by another model if the chosen one fails; the model that wrote it is shown.
Consensus. In a Consensus session several models answer the same question, review each other's answers, and one model writes a synthesis. The result shows where the models agree and where they disagree. A confidence indicator shows how far the models agree. It is not the probability that the answer is correct.
ewpire's outputs are AI-generated content and estimates, not professional advice, and they may be wrong. Figures in an output, including market-size estimates, are unverified. ewpire does not make any automated decision producing legal or similarly significant effects concerning you within the meaning of Article 22 GDPR. ewpire screens image, video and music prompts for the content listed in Section 5; model providers apply their own safety filters to all requests.
3. AI-generated outputs are marked (Art. 50(2))
Everything ewpire generates – answers, code and sites, images, videos, music and, where available, AI Visibility content – is artificially generated. What is marked today:
| Output | Visible marking | Machine-readable marking |
|---|---|---|
| Images (PNG, JPEG, WebP) | "AI-generated" label on every image in the product | where the model's maker embeds signed provenance data (C2PA), the file is kept exactly as delivered so the signature still validates; otherwise embedded metadata: IPTC digital source type "trainedAlgorithmicMedia" (XMP) naming the tool and the model (PNG also text fields); downloads carry the header X-AI-Generated: true |
| Videos (MP4) | "AI-generated" label on every item in the product | kept as delivered where signed by the model's maker; otherwise an XMP box with the same digital source type and the model; downloads carry the header X-AI-Generated: true |
| Images in GIF, videos in WebM | "AI-generated" label on every item in the product | download header X-AI-Generated: true; no embedded marker yet |
| Music tracks (MP3) | "AI-generated" label on every track in the product and its player | where the provider delivers the track with its own watermark or provenance data, the file is kept as delivered; otherwise ewpire adds an ID3 tag stating that the track is AI-generated and the model; downloads carry the header X-AI-Generated: true |
| Files made from a track (WAV copies, stems, music videos, cover images) | shown under the track they come from, which carries the "AI-generated" label; every file name says "AI-generated" and the model | an embedded tag or comment stating that the file is AI-generated and the model (cover images: the PNG marking above); downloads carry the header X-AI-Generated: true |
| Chat and Consensus answers | shown with the model that wrote them | plain text answers carry no embedded marker |
| Code answers and versions | the model is not named yet | exported and published sites carry an "AI-generated" meta tag |
Images in GIF and videos in WebM are labelled in the product and on download; they do not yet carry an embedded marker. A frame you save from a video as an image is not marked yet. Embedded metadata is lost when a file is re-saved, converted or screenshotted; where the model's maker adds an invisible watermark, that watermark can survive such changes. Check a file: at ewpire.com/check you can see whether a file carries ewpire's label or signed provenance data; the file is read in your browser and not uploaded. Do not remove the label or the provenance data from a file you publish (Terms of Service §5.9).
4. Your duties when you publish AI-generated content (Art. 50(4))
If you publish or distribute content that ewpire helped generate – in particular images, audio, or video that constitute a deepfake, or AI-generated text published to inform the public on matters of public interest – you must disclose that the content is artificially generated or manipulated, as Art. 50(4) requires. Keep the embedded metadata of downloaded images, and the watermark and provenance data of downloaded music tracks, intact where you can. A song whose vocals imitate a real, identifiable singer would be a deepfake; the Terms of Service prohibit creating one (§5.13(c)). This duty also applies to referral partners who promote ewpire using AI-generated content (see the Referral Programme Terms).
Where the API is available to you and you build your own application on it, you must give that application's users the Article 50 disclosures that apply to it.
5. Content safety for images, video and music
Every image and video prompt is screened before any credit is held. ewpire refuses prompts that seek sexual content involving minors, intimate or sexual imagery of a real person, or a deepfake or face swap of a real person, including of a person shown in an uploaded photo. Music prompts are screened the same way for requests that name or imitate a real singer or artist, ask for a real person's voice, or reuse the lyrics of an existing song; Music does not accept voice or audio uploads. This screening cannot catch everything. The model providers apply their own safety filters in addition; a request they refuse is not charged.
Some models are marked "third-party access" in the model catalog. Before your first request to such a model we tell you what that means for your content and ask you to accept it (Terms of Service §8.2).
6. Editorial responsibility
You are responsible for any AI-generated output you choose to publish, deploy, or rely on. Outputs may contain errors; review them before use.