Platform labels are a starting point
TikTok asks creators to label AI-generated content. Since May 2024 it has also read C2PA Content Credentials to label AI-generated images and videos made on some other platforms automatically, and it attaches Content Credentials to some AI-generated content made on TikTok. A label and an independent detector answer different questions, so keep both in view, and treat a missing label as inconclusive.
Source: TikTok: advancing AI transparency (May 2024) (opens in a new tab)
Use the right material
A TikTok post URL is a platform page, not a media file, and downloaded or screen-recorded copies are re-encoded. If you have the right to analyze a clip, the creator's original file is the strongest evidence. For a still image, use the original rather than a screenshot that mixes the picture with captions and interface elements. Image Evidence does not download platform videos, log in or inspect private accounts.
Know what frame analysis misses
Short clips can include rapid cuts, overlays, filters and mixed sources. Frame-based checks sample a limited number of frames rather than evaluating every moment, and they do not authenticate speech, match faces or determine whether a real person said particular words. A single averaged score can hide variation across the timeline, so look for results that report their coverage rather than assuming continuous analysis.
Ask about the specific claim
Is the clip presented as comedy, an illustration, a product demonstration or news? Whether pixels were generated does not resolve whether the caption is accurate. A camera recording can still stage an event, and a generated illustration can be openly disclosed. Review the original account's context and look for independent corroboration when the claim matters. Do not turn a tool result into a judgment of somebody's identity or character.
Treat results as one clue
Metadata checks can return useful clues, but a C2PA structure needs signature validation before its issuer can be trusted, and a SynthID check requires Google's own tools. Image Evidence has no published TikTok benchmark and does not train on user content. There is no watermark removal, evasion or Humanizer functionality.
How can you tell if a TikTok video is AI-generated?
Check for TikTok's AI-generated label and the creator's caption, then look at the account's other posts and any linked sources. For realistic footage of real events, compare it with independent reporting. Visual glitches can prompt questions, but polished AI video and heavily edited real video can both look unusual.
Does TikTok automatically label AI content?
Partly. TikTok labels content made with its own AI effects and reads C2PA Content Credentials from some other platforms to apply labels automatically. Content without those signals depends on creators labeling it, so an unlabeled video is not proof that it is real.
Source: TikTok AI-generated content guidance (opens in a new tab)
Can AI detectors check TikTok videos?
Some can analyze an uploaded clip by sampling frames, but downloaded TikTok copies are compressed, which weakens both metadata and pixel evidence. Results are best treated as one signal next to the label, the account's history and independent reporting.