TikTok AI label triggers mapped to the Sume tools that cause them

Each trigger AuditSocials lists for a TikTok AI label has a matching Sume surface, except voice cloning, which is not in Sume's API. The summary (third-party, read 2026-10-02) names synthetic faces and face swaps, AI voice cloning, and photorealistic AI backgrounds and product images. The table below matches them to what you would call.

Which Sume surface produces each trigger?

Read the right-hand column as an inventory of where a label decision is needed in your pipeline, not as legal advice.

Does a talking avatar count as a synthetic face?

The summary says synthetic faces need a label, and an Avatar 1.0 video is built around a generated face. Plan on labeling it.

What about the voice?

The avatar voice comes with the talking-video job. Sume's API does not clone a voice from a sample; that is an app feature. So the voice-cloning trigger does not arise from the API, but the label question for a synthetic voice on a synthetic face is still yours to answer.

What should the pipeline record?

Per clip, record which row of the table applied. A clip can hit two rows, such as an avatar in a generated background. If any row applies, the answer is a label. Re-read TikTok's own policy before you finalize the rule, since this list comes from one summary.

Which trigger is easiest to miss?

Backgrounds. A talking-avatar clip has an obvious synthetic face, so people remember to label it. A real person filmed in front of a generated scene looks ordinary in the edit bay, and the summary lists photorealistic AI backgrounds as a trigger all the same.

Review the prompt as well as the output. If the prompt asked for a photoreal place, a product on a surface, or a crowd, the clip belongs in the labeled group, whatever the final frame looks like.