EU AI icon with a "voices generated with" label on avatar video

On the Commission's EU icons page, the use case example for the basic icon is a deepfake video carrying the text label "voices generated with" followed by the icon. On Sume you can author that text line yourself as a caption cue on an avatar video, because cues are burned in as written and skip speech-to-text.

This is a layout how-to, not legal advice. The Commission says the icons are optional, that using them does not establish legal compliance, and that deployers stay responsible for meeting Article 50 of the AI Act. Source facts read 2026-10-01.

What does the Commission say about the text label?

The page states that user testing improved performance across all measures when the basic icon was accompanied by a text label (its example is "modified"). It lists the basic icon for when AI was involved in creating deepfake content, or when a custom text label is used. Its accessibility advice is to keep the label in plain language and avoid abbreviations other than "AI".

How do I author the text line on an avatar video?

Generate the talking video from a script with Avatar videos, then send the finished clip to POST /v1/video-captions with cues. Each cue has text, start and end in seconds. The captions doc marks cues as an authored overlay that skips speech-to-text, so your exact words are burned in.

const res = await fetch("https://api.sume.com/v1/video-captions", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.SUME_API_KEY,
    "Content-Type": "application/json",
    "Idempotency-Key": "ai-voice-label-001",
  },
  body: JSON.stringify({
    video_url: "https://media.sume.com/your-avatar-clip.mp4",
    cues: [{ text: "Voices generated with AI", start: 0, end: 6 }],
  }),
});
console.log(res.status, await res.json());

Does a caption cue draw the EU icon itself?

No. Cues carry text only. The icon is a separate graphic that you download from the Commission, so the cue covers the words and a still-over-video step covers the graphic; see burning the icon into the first frame. Because the Commission asks for no intervening overlay elements, also read keeping captions off the label before you add speech captions to the same clip.

What if the clip has no speech?

Speech-to-captions needs audible speech; the docs say a silent clip fails as caption_no_speech. Authored cues avoid that path, which is why the label line works on a silent opener too. See silent clips and overlay cues.

What wording suits a synthetic voice?

The Commission's own example text is "voices generated with" followed by the icon, so a cue such as "Voices generated with AI" mirrors it, and the accessibility advice is plain wording with "AI" as the only abbreviation. When the whole clip is synthetic, compare the icon variants in partially modified vs fully generated.