AI Transparency

1.The visible “AI” label

Every video exported from Talking Photos, on the free plan and on PRO, shows a small badge with the capital letters AI in the top-right corner, from the first to the last frame. The letters are white on a dark plate with a white ring, so the badge stays readable on light and dark pictures, and it is sized relative to the frame (about 6% of its shorter side) so it stays readable when the video is shown small on a phone. It is drawn into the picture itself, so it travels with the video wherever it is shared. The same badge appears in the preview in the editor.

What it means: the video was generated with AI. A real photo was animated by software and the voices are synthetic, so the people or animals in it never actually said or did what you see and hear.

The design and placement follow the European Commission’s Code of Practice on Transparency of AI-Generated Content (Section 2, Measures 1.1 and 1.2): the capital acronym “AI” as the main element, letters of equal height, high contrast, top-right corner with nothing drawn over it, visible for the whole video. On the free plan a separate “Talking Photos” badge sits in the top-left corner; that one is our logo, not the AI disclosure.

2.Machine-readable marks

Under Article 50(2) of the EU AI Act, AI-generated video must also be marked in a way software can detect. Every exported MP4 carries three marks:

• C2PA Content Credentials. A digitally signed C2PA manifest embedded in the file, signed (ES256) by My Little Indie Studio with our own certificate chain (root “Talking Photos C2PA Root CA”) and time-stamped by DigiCert’s public RFC 3161 time-stamp service (only a hash of the manifest is sent to it, never the video or any data about you). It names the claim generator (“Talking Photos” with its version), records a c2pa.created action with the IPTC digital source type compositeWithTrainedAlgorithmicMedia — a real photo animated by AI with a synthetic voice — and a description saying the video was generated with AI by Talking Photos, and declares the audio watermark in the assertion app.talkingphotos.watermark (algorithm tp_wm_v1, medium audio). Any change to the signed content makes the signature check fail.
• MP4 metadata. The comment and description tags read “AI-generated with Talking Photos (talkingphotos.app). Contains C2PA Content Credentials.”
• Audio watermark. An imperceptible spread-spectrum mark in the soundtrack (tiny energy changes between 0.5 and 8 kHz, far below the sound itself) carries only a random 32-bit code for the export. In our tests it was still detected after re-encoding to AAC 128 kbit/s and to MP3 at 128 and 64 kbit/s, resampling to 44.1 kHz, volume changes of ±6 dB and cutting the video down to a 5–10 second excerpt, and it gave no false alarm on 96 unmarked clips. From a whole video the check reads the export date; from short excerpts it usually reports only that the watermark is present.

None of the marks contains your name, email address, account id or IP address. The watermark code is random; our server maps it only to the export date (see the Privacy Policy).

The marks complement each other: many platforms strip metadata when they re-encode uploads, while the watermark lives in the sound. No technique is perfect, though: re-recording the video through a microphone, heavy audio editing (pitch or speed changes, added noise), removing the soundtrack, filming a screen or deliberate tampering can defeat the machine-readable marks. A video without our marks is therefore not proof that it is real or that it was not made with Talking Photos.

3.How to check a video

• Open our verify page (“Check a video” in the footer) and choose the video file — no account needed. It reports the Content Credentials (present, valid, signer, generator, date, source type) and the audio watermark (detected, confidence, export date) separately. Tools can use the same check through POST /api/v1/verify. Files up to 100 MB; up to 10 checks per IP address per 10 minutes.
• Check the Content Credentials independently with any C2PA tool, for example contentcredentials.org/verify. Because we sign with our own certificate authority, such tools show our signer as not on the C2PA trust list; the signer name “My Little Indie Studio” tells you the credentials are ours.

A checked file is analysed and deleted immediately; we do not keep it or log its content. IP addresses are used only in memory to apply the check limit and are never stored.

4.When you share or publish

Keep the label and the marks: our Terms forbid removing, hiding or tampering with them. You may edit, trim or re-upload your videos, as long as the “AI” label stays visible in the corner whenever generated footage is on screen.

If you publish videos as part of a business or professional activity, Article 50(4) of the EU AI Act makes you responsible for disclosing that the content is AI-generated; the label in the video is designed to meet that duty, so keep it. For evidently artistic, creative, satirical or fictional work the disclosure may be placed so it does not spoil the work, but it must still be clear when people first see the video.

5.Report a missing or incorrect label

Write to info@talkingphotos.app if you notice that:

• a video made with Talking Photos has no “AI” label, or the label was removed or covered;
• the verify page gives a wrong result, or Content Credentials or the watermark are missing from a fresh export;
• a video that was not made with Talking Photos is shown with our marks;
• a video was made from your photo, or of you or your child, without consent.

Please include a link to the video or the file and a short description. We look into every report, correct our marking if it is at fault and act on misuse under our Terms.