EU AI icon: Partially AI-Modified or Fully AI-Generated?
The Commission's icons page gives a face swap in an authentic photo as its Partially AI-Modified example. What that means for a Sume face-swapped video.
By the Commission's own examples, a face swap in existing footage looks like the Partially AI-Modified case, and a clip with no human-made elements apart from prompting looks like the Fully AI-Generated case. The page's example for Partially AI-Modified is a face in an authentic photograph swapped with a politician's face. Sume ships no icon, so you add it, and this is not legal advice.
Icon definitions are from the Commission's page; face swap and avatar facts from Sume's Face swap docs. Both read 2026-09-30.
What are the three icon variants?
| Variant | When to use | Example on the page |
|---|---|---|
| Basic | AI was involved in creating the deepfake, or a custom text label or second layer is used | Video with the text label "voices generated with" and the basic icon |
| Fully AI-Generated | The entire content is generated by AI, with no human-created elements or editorial control apart from prompting | Fully AI-generated deepfake videos with politicians or fictional events |
| Partially AI-Modified | Pre-existing, human-made content was partially modified with AI, turning it into a deepfake | A face in an authentic photograph is swapped with a face of a politician |
Where does a Sume face swap fall?
Sume's face swap applies a ready avatar's face onto a public HTTPS source video; the docs say the Beta targets clips of about 4 to 15 seconds with usable audio. The footage already existed and the face was changed, which matches the shape of the Partially AI-Modified example. Whether your clip is a deepfake at all, and which variant applies, is a legal question the page leaves to deployers.
The avatar itself can start from a prompt, structured traits, or a reference image, so the source of the face varies. That does not change the clip's label by itself; the Commission's test is the human-made elements in the final video. See face swap a video with a photo for the workflow.
And a clip generated from scratch?
A clip made only from a prompt, with no human-made footage in it, is the Fully AI-Generated description. If you add real footage, a human voice track or editing of your own, the "no human-created elements" wording may no longer fit, so read the page's wording against your own clip.
Does the icon replace the platform disclosure?
The Commission says the icons are optional, the Article 50 labelling requirements are not, and deployers remain responsible for meeting them. Platforms have their own settings; for YouTube see disclosing a face-swapped video of a real person. To embed an icon in the file, see burning the EU icon into the MP4.
Sources
Related posts
More in Use cases
- Facebook Feed video ad 4:5 at 1440x1800: render it with Timeline
Meta's Facebook Feed video spec lists 4:5 at 1440 x 1800. No Sume video model lists 4:5, but a Timeline 1.0 render can output 1440x1800. Steps and limits.
- Facebook Feed video captions: burn them onto a clip with no speech
Meta marks captions optional but recommended for Facebook Feed video ads. Sume's caption job fails on a silent clip unless you pass your own cues.
- Facebook Feed video ad file type: MP4, MOV or GIF from Sume?
Meta lists MP4, MOV or GIF for Facebook Feed video ads. Sume Timeline, video-filter and video-trim each return an MP4, so MP4 is the format to plan around.
- Flow Music edits a song by section: how to do it by API
Google Flow Music edits songs by section; its Lyria API page says multi-prompt editing is unsupported. On Sume, cut two takes and join the parts you keep.
Written by Sume