Meta AI disclosure: which Sume outputs count, video and audio

Meta's manipulated media policy asks people to use its AI-disclosure tool when they post organic photorealistic video or realistic audio that was digitally created or altered. Penalties are possible, and Meta may add informative labels to high-risk deceptive content. Photorealistic generated video, talking avatars and realistic voice or music generally land in scope; clearly stylised animation is the grey area.

The policy points

This is the organic-content policy. Paid ads have a separate path in Ads Manager, covered in other posts.

Sorting Sume outputs

The table is a working guide, not a legal ruling. It applies the policy's wording to the output types that Sume documents.

Make it a workflow step

Put the disclosure decision at the same moment as the upload. A per-clip sheet with the job ID, the output type and a yes or no for the toggle makes the choice visible. The video API returns a job ID on submit, and the avatar docs and Music Router docs cover the other two output types.

When a clip mixes real footage with a generated segment, consider disclosing the whole post. The policy summary we read says nothing about proportions, so label when unsure.

clip,job_id,output_type,meta_ai_toggle
spring-hero-9x16.mp4,<job id>,photorealistic video,yes
spring-theme.wav,<job id>,realistic audio,yes
spring-cartoon-cut.mp4,<job id>,stylised animation,review

Caveats

Disclosure is cheap compared with a takedown or a trust problem. Leaning towards labelling is the lower-risk default.

• The policy page can change; check the date you read it.
• This post does not cover ads, which carry their own labels.
• If your output type is not in the table, ask which part a viewer would take as real.

Edge cases

Voice is the case people miss. A realistic synthetic voice over a real photo is still realistic audio that was digitally created, and falls under the same wording. A music bed generated for a clip belongs on the same list if it is presented as a real recording.

Another edge is partial edits. If you trim, caption or crop your own footage, you have not added synthetic content, and the table above says so. If you replace a face, a voice or a background with generated material, you have.

When two people disagree on a clip, ask which one would feel misled if they learned how it was made. Label if the answer is anyone.