AI Transparency

1The video is made by machine#

Every part of your video is generated: the script from your story, the characters from your photos, each scene, the narration and the subtitles. No artist draws it and no actor reads it. One person runs the service, looks at what fails or has to be put right and answers the support address; apart from that, nobody sits between your story and the video.

The models are a text and image family from Google for the writing and the drawings, Grok Imagine from xAI (SpaceXAI, LLC) for putting the drawings into motion (a scene it could not animate at all is shown as its drawing, held still or with a slow camera move), and from ElevenLabs a voice model for the narration and a music model for the music under your story. The cards at either end - an opening for each occasion and one ending - were generated with Grok Imagine and then edited by us, and are ours. When Google's reading and writing models are all out of capacity at once, Grok, also from xAI, stands in for them and does that part of the work. Which exact model does which job changes as they improve; what the privacy policy promises about your photos does not change with it.

2It is marked as synthetic#

The video ends with the European Union's “AI” icon in the corner of its closing card, the sign the European Commission published for content made with AI, the same in every language. The MP4 file carries the same fact for software to read: signed Content Credentials, in the open C2PA standard, that record the video as generated by AI, which anyone can read at contentcredentials.org/verify, and a plain line in its metadata. Inside our credentials are those the models' makers signed on each scene's clip, or on its drawing where the drawing is shown instead, kept as they were, and the picture of your video shown on its page and in the email carries credentials of its own. We sign the credentials with our own certificate, which is not on the list of signers that site trusts, so it shows them as coming from a source it does not recognise. Marking what our service makes is our duty under article 50 of the AI Act (Regulation (EU) 2024/1689), as the provider of the service that generates it, and these are how we mark it. The credentials and the metadata line travel with the file rather than with the page it was downloaded from, but many of the places a video is posted strip metadata, and the pixels themselves carry no invisible watermark of ours. We do not hide that a machine made it, and the terms ask you not to remove the credentials or the metadata line or change them to say anything else, and not to present the video as a recording of something that happened.

3What your photos are used for, and what they are never used for#

A photo is read to learn what a person looks like, their apparent age included, and is used to draw a character who resembles them, on the figure of an adult, a teenager or a child. Your photographs never appear in the video; the site shows them only on your own screen as you add them, and crops of the people in them on your own video's page when we ask you who is who. They are never published, never shown to another customer, and never used to train a model of ours; besides you, they may be seen by the one person who runs this service, when something fails or has to be put right, or to answer a request about them, and, for up to 55 days, by authorised staff of Google checking for abuse; the providers that receive them are named in the privacy policy, with what each keeps and for how long.

Faces are not compared with each other. The system asks which character in your story a person is, and when it cannot tell, it asks you instead of matching one photograph against another. When your story names someone who is in none of your photos, or too small in them to draw, it asks you for a photo of them rather than inventing a face.

4What it gets wrong#

A generated video is an approximation. A character may resemble you the way a good caricature does rather than a photograph; hands, small objects and text inside a scene come out wrong more often than faces do; and a scene may read the story more literally than you meant it. We redraw what falls below our own bar, though a flaw can still get through, and the price is the same whether the machine needed one attempt or five, but none of that makes the result a photograph of your life.

5What it refuses#

The models refuse some content outright, and so do we: the terms list what you must not send. Nothing reads what a story shows before you pay. After you pay, a story that asks to show a child in a sexual or abusive situation comes back to you, with the passage named whenever it can be found word for word in your text, so you can change it and the video goes on, and so does a story that mentions more than five people, with the people counted; a refusal by a model's own safety rules cannot be edited away, and then the video is not made and the payment comes back at the 48 hours. A refusal you think is wrong reaches a person at support@amovi.eu.