Meta's AI info label: signals or self-disclosure, and a trimmed clip

Meta says it labels AI content when it detects industry-standard AI image indicators, or when people disclose it themselves. The first route depends on metadata or watermarks that must survive your pipeline; the second does not. Sume documents trim, filter and timeline outputs as new MP4 files but I found no statement that they carry provenance metadata through, so treat self-disclosure at upload as the dependable route.

What Meta's post says

Meta's labeling post, first published April 2024 and read 2026-10-04, says labels come from industry-standard indicators or self-disclosure, that content only edited with AI has its label moved to the post menu (an update it dates to September 2024), and that Meta keeps content up to add labels unless it breaks other policies. It is a 2024 statement; check Meta's current help pages before you ship a dated policy claim.

Check the file you will upload

Use video inspect on the final file: it reports container, codecs, has_audio and size. It does not read C2PA manifests. For range cuts, video trim with precision: exact re-encodes the video, and keyframe copies the streams; neither is documented as a metadata guarantee.

So the plan is simple: keep a list of generated clips, and make the self-disclosure at upload for every file that contains one.