Meta skips AI labels: no significant edit, no real-looking human

Meta's news post says that if its generative AI creative tools do not produce significant edits to the image or video and do not include a photorealistic human, it will not apply any AI label. That sentence is about Meta's own tools; a Sume trim, crop or dim is not a Meta tool, so it does not inherit that exemption.

The wording is from Meta's post (read 2026-10-10). I am not extending it to third-party media tools, because the post treats them separately.

The exemption in Meta's words

The post says Meta applies AI info labels when ads are created or significantly edited with its generative AI creative features. It adds that if the tools do not produce significant edits to the image or video and do not include a photorealistic human, it will not apply any AI labels. For where labels appear, it says behind the three-dot menu, or next to the Sponsored label when the tools result in the inclusion of an AI-generated photorealistic human.

Why this is about Meta's tools

The exemption sits in the paragraph about Meta's own features. The post then says separately that Meta will begin automatically detecting ads created or edited using third-party AI tools through industry-standard signals. Those are different mechanisms, and the exemption text is not repeated for the second one.

What trim, crop and dim actually do

The Sume docs describe video trim, video filter and Timeline as plain ffmpeg on a worker, with no provider inference. A trim cuts a range and returns a new MP4 for $0.02. A crop op takes fractions of the frame. Dim multiplies luma by an amount above zero and up to one. None adds a person, a scene or a sound. If your source is real footage, these edits leave it real.

If your source is a generated clip, the edit does not undo that. A trimmed generated clip is still generated. The significance test in Meta's sentence is about the edits Meta's tools make, but a viewer or reviewer looking at the ad is looking at the content.

A cautious reading

Use the exemption sentence as a statement about Meta's tools, not as a rule for the file you upload. If an ad includes a photorealistic human who never existed, Meta's own post says the label can sit next to Sponsored, and a reasonable team would expect that placement for any realistic generated person, whatever tool made it. When you plan a campaign, decide on honest copy, then let the platform apply whatever labels it applies.

It is easy to turn a single sentence into a rule it was never meant to be. Meta wrote that sentence to describe its own creative tools, such as generating or expanding an image inside Ads Manager. It did not write it as a general safe harbour for edits made anywhere else. If you read it that way, you will make decisions that Meta's own page does not support.

A safer working rule is to ask two questions per ad. Did a model create any of the picture or sound? And does the ad show a person who looks real but is not? If either answer is yes, expect that a label is possible and brief the client. If both answers are no, you have filmed footage with ordinary edits, and the page's concern about generative features is not triggered by anything you did.