Detection
Videos, GIFs and thumbnails: what is actually checked
Video frames are never analysed. Containers are scanned for Content Credentials, posters and YouTube thumbnails are checked as images, and animated images get provenance only.
Updated
The card is honest about what it looked at. The muted line under the detector rows says, for example, "Checked the video's thumbnail" or "Video frames not analysed; container and poster only".
Video files (MP4, WebM)
Our API fetches the first 4 MB of the file and scans the container for a Content Credentials manifest (MP4 and other ISOBMFF containers; WebM and Matroska are out of scope). The poster image, if the player has one, is checked as an image with every image detector. Frames are not decoded.
YouTube
Thumbnails on the home page, search, sidebars, Shorts, channel grids and end screens are checked as images, labelled "video thumbnail". On watch pages and in embeds the player is cross-origin and cannot be captured, so the extension derives the video's public thumbnail from its id and checks that. See YouTube Shorts and watch pages.
Other players
Vimeo, TikTok, Twitch and Facebook players use their poster or Open Graph image. Same limitation: poster only.
GIFs and animated images
Animated GIF, WebP and APNG files get provenance checks (Content Credentials in XMP, EXIF and IPTC where present, and the watermark check) but no classifier or recheck by default; the card says "Animated image: provenance only". A Pro recheck decodes the first frame server-side when the model accepts it.
Same-origin video frames (Pro)
A Pro recheck can capture a frame from a video served by the same origin as the page. It never attempts capture inside cross-origin YouTube, Vimeo, TikTok, Twitch or Facebook players.
Audio
Not analysed anywhere. The popup says so.