AI info on the post or in the menu? Generated vs edited clips

On Meta's apps, content detected as generated by an AI tool keeps a visible AI info label, while content detected as only modified or edited by AI moves the label into the post's menu. Meta announced that change on September 12, 2024 (read 2026-10-03). So the question for your clips is whether Meta would see them as generated or merely edited.

Meta does not publish a per-tool list, so the mapping below from Sume job types to those two buckets is our reading and nothing more.

What did Meta change in 2024?

Meta's labeling newsroom post gives a short timeline. On July 1, 2024 "Made with AI" was renamed "AI info", because indicators of minor retouching were over-labeling real photos. On September 12, 2024, content detected as only modified or edited by AI had its label moved to the post's menu. The label is still displayed for content detected as generated by an AI tool, and Meta shares whether the label is due to industry-shared signals or self-disclosure (read 2026-10-03).

The same post says AI info labels apply when Meta detects industry standard AI indicators or when people self-disclose. Meta may also add a more prominent label for high-risk deception.

How would Sume job types map?

Sume's models either generate media from nothing, or transform media you supply. The table below sorts a few job types. Treat the right column as a guess about detection, not a promise.

Does the bucket depend on Sume at all?

Mostly no. Detection is Meta's, and Meta says it relies on industry standard indicators. Sume's docs describe job records, artifacts on media.sume.com and usage, and do not claim that outputs carry C2PA credentials or watermarks. So you should not expect Sume to provide a signal that makes Meta label or not label a clip.

What you control is self-disclosure. If a clip is generated, add the label yourself when you publish. It costs nothing and means the label is not left to detection. Our post on what Sume adds to AI video covers the file side, and checking a delivered file shows how to test what survives.

What about a clip that is both generated and edited?

A generated avatar video that you then trim and caption is still a generated clip. The edit steps do not turn it into an edited one. If you trim a real creator video and add captions, nothing generative happened, and there is no reason for an AI label.

Mixed cases are where records matter. For each final file, note the chain of jobs that produced it, in order, and which step created the people or scenes. A small table in your own tracker is enough.

• Final file URL and the date you published it.
• First generative step, if any, with job id.
• Every later edit step with job id.
• Whether you self-disclosed on Instagram, and where.

What should you do before publishing?

Decide by what the viewer would think is real. If a person on screen is not a real person, or a scene did not happen, label it, whichever bucket Meta would put it in. The menu placement for edited content is a Meta product decision about prominence; it is not a loophole, and Meta says the experience may vary by region and can change.

For ads, see what Meta's help page says about video ads: the ad labels use a different surface, About this ad, and are documented separately.

What does the menu placement change for viewers?

A label in the post menu is something a viewer has to look for. A label on the post is something they cannot miss. That difference is the whole point of Meta's September 2024 change: content only modified or edited by AI is judged less likely to mislead, so its label moved out of the way.

For a brand, that is not permission to rely on the quieter placement. A viewer who opens the menu on a clip you made and finds an AI info label you did not mention may feel misled, whatever the bucket. If your clip contains a generated face, the cleaner approach is to say so in the caption or on the picture, and let the platform label add to that rather than replace it.