2. What AI does in our work
Restoring the original. We remove scratches, stains and creases, recover sharpness and, where needed, colour. The face stays the face of your person: we do not change features, age or expression.
Artistic reconstruction. If the photo shows a real scene — a room, a yard, a street — we keep it and only restore it. If the background is empty (a studio portrait) or lost, AI paints a plausible setting of the era and region. A Description, if you add one, helps us recreate details more accurately.
Living Presence technology (2.5D). From a flat photograph we build a three-dimensional scene — the camera glides slowly through the frame. Light movement — just enough to give the scene depth, and no more.
Assembling the Film. We edit the restored photos and movement into a Film or Reel with music and sound design: horizontal 4K for Smart TV, vertical for Viber, a short version for Stories/Reels — depending on the Package.
Captions and titles. Text models help us draft captions from your Descriptions — names, years, places. We check every line.
There is a person at every step: we choose which generated material goes into the work, and we review every frame before delivery.
3. What AI does not do, and what we never do
- We do not synthesise a voice — neither your person's nor a "similar" one.
- We do not make people "speak": no lip movement, no invented speech.
- We do not work with photos of living people without their consent (except you yourself). For children — only if you are their parent or guardian, or have that person's consent.
- We do not create content meant to deceive: no "evidence", forged documents, invented events or disinformation.
- We do not create sexual or nude content.
- We do not depict children outside a family context.
- We do not claim to "bring back" or "revive" a person. We bring their memory to life — which is a different thing.
If a scene conflicts with these rules or our Content Rules, we decline the Order and refund the part not performed.
4. What "artistic reconstruction" means, and how accurate it is
An Artistic Reconstruction is an image restored by AI: facial features come from the original, while colours, fine details, damaged areas and — where there was none or it was lost — the background are recreated by AI. "Recreated" is the key word.
From the original — the face. We restore it from your photograph and check that it has stayed itself (section 8).
Recreated — colours (a black-and-white photo did not keep the real ones), fine details the original does not show, lost areas, and the background where there was none. We do not know what colour your great-grandmother's dress really was; we know what dresses of her time looked like. Those are different things, and we do not pretend otherwise.
What we check. Anachronisms — objects, clothing or technology that could not have existed at that time and place. Language — signs and lettering in a scene must fit the era and region, and captions must be correct Ukrainian (or English in the EN version). The face — so that restoration does not "improve" a person beyond recognition.
What may still be inaccurate. The cut of a dress, the façade of a house, the colour of a wall, the shape of a cup; sometimes details the family remembers differently. The Film is a work of art, not a historical record and not evidence of how things were. Please do not use it as proof, and do not present the scenes as authentic photographs.
If you spot an error — write to [email protected] within 14 days of delivery: which scene and what is wrong. One revision round is included in every Package and covers defects within our control: artefacts, errors in captions or titles, obvious distortion of a face, a mismatch with the Package ordered. Recreated details that do not contradict the original or your Description are not defects.
5. How we label AI content
For now the label is where the Deliverables are watched most — on the Delivery Page:
- a line at the end of the Delivery Page saying that this is an artistic restoration made with artificial intelligence;
- on the Studio's social media — a line in the caption plus the platform's "made with AI" label switched on (we publish only under your Marketing Release);
- we do not remove the marks that model providers add to files.
The Film, the downloadable files and the Photo Book do not yet carry a separate visible label. When they do, we will update this page.
Please say it is AI when you share. Under our Terms of Service you are free to share the Film with your family and on your personal pages — provided the "made with AI" label is switched on when you post on public platforms, or the caption says the photo was restored with AI. That is how relatives who see the Film without you will understand what they are looking at.
Label texts / Тексти маркування — exact strings
Where EN UA Delivery Page Artistic restoration made with the latest advances in artificial intelligence and Vidlunnya's living-presence technology, with a person involved at every stage. Художня реставрація, зроблена за допомогою найбільших досягнень штучного інтелекту та технології живої присутності студії «Відлуння», де людина бере участь у кожному етапі створення. Social media caption (+ platform label) Made with AI: the photo was restored from the original. Створено за допомогою ШІ: світлину відновлено з оригінального фото.
6. Your rights and our obligations under the law
EU Artificial Intelligence Act, Article 50(4). An image or video that resembles a real person and could falsely appear authentic is what the Act calls a "deep fake"; whoever creates it for others must disclose that the content has been artificially generated or manipulated, no later than the first exposure. We are such a party, and our Deliverables are used in the EU. The law allows an unobtrusive disclosure for artistic works. For now we disclose it on the Delivery Page, where the Deliverables are first shown; we are preparing a label in the Film itself and in the files. The obligation applies from 2 August 2026; the European Commission's Guidelines of 20 July 2026 recommend a clear visual label at first exposure, an opening disclaimer for video and a persistent label for images, and we are working towards that.
Ukraine. There is no dedicated AI law yet. We follow the voluntary AI Code of Conduct of the Ministry of Digital Transformation (December 2024), which recommends labelling AI content.
Our providers' terms. OpenAI's Sharing & Publication Policy requires AI content to be indicated "in a way no user could reasonably miss or misunderstand", every generation to be reviewed before sharing, and the content to be attributed to its creator. We do this for every Order.
Platform rules. Meta (Instagram, Facebook) asks for the "AI info" label on photorealistic content that was created or altered. TikTok requires the "AI-generated" label and strictly limits synthetic media showing private individuals and minors — which is why we publish "before/after" only under your Release and never publish children. YouTube requires the "altered or synthetic content" disclosure when a real person is shown where they were not. When you share the Film yourself, please do the same.
Your rights. You may ask us which parts of the Film were generated; have a scene removed or replaced within the revision round; complain to us or to the supervisory authority of your country. The Deliverables are yours for personal, non-commercial use.
7. Which technologies we use
- Images: OpenAI models via API (restoration and frame generation).
- Faces: face-box detection (InsightFace) runs on our own server; images are not sent anywhere.
- Depth, sharpening and Film assembly: the depth model (Depth Anything), the sharpening model (Swin2SR) and the renderer itself run on rented Runpod GPUs or in the operator's browser. Only the frames needed for rendering reach the GPUs, and only for the duration of processing — the finished video goes straight back to our EU storage.
- Editing and rendering: our own pipeline for assembling the Film, titles and the Photo Book layout.
- Hosting: server and database — Hetzner, Helsinki, Finland (Hetzner, EEA); files and video delivery — Cloudflare (Cloudflare R2, bucket with European Union jurisdiction (files are stored only in the EU; the CDN serves them from the nearest edge)).
Under our contracts, your photos are not used to train models: OpenAI does not train on API data and keeps service logs for up to 30 days. Face detection involves no third-party providers; depth mapping, sharpening and rendering run on rented Runpod GPUs only for the duration of processing — no model training and no retention of frames. The full list of sub-processors and retention periods is in our Privacy Policy.
8. The face check
To make sure a restored face is still the face of your person, we may run a temporary automated Similarity Check. For a few seconds the software computes a numerical representation of facial features in the original and in the restoration, compares them — and destroys it immediately. Those numbers are never stored, never compared with any database, never used to identify or authenticate anyone, never sold and never used for training. You give separate consent at checkout; if you would rather not have the automated check, write to us and a person will check the likeness instead. More in our Privacy Policy.
9. How to report a problem
Write to [email protected] or on Telegram at @vidlunnyastudio. We acknowledge receipt within 2 working days and give a substantive reply within 5 working days. If it is a defect, we fix it within the revision round; if someone has removed a label or the Film is being used to deceive, we help you document it and, where possible, contact the platform; if you believe we have broken the law, you may contact the supervisory authority of your country (the list for personal-data matters is in our Privacy Policy).
10. Changes
We update this page when technologies, providers or the law change. We announce material changes on the Site and by e-mail 14 days in advance. Orders already paid for are governed by the version in force at the time of payment.