YouTube AI label: beauty filters, color skip it; regens may not

YouTube's help page names three kinds of edit that need no altered-or-synthetic disclosure: beauty filters, color adjustment and audio repair, which it calls minor edits. A regenerated scene is not on that list. If it shows something realistic that did not happen, or alters real footage of an event or place, the four triggers apply.

Exempt versus not

Sume's non-generative tools

Several Sume media tools are plain ffmpeg jobs. Video trim cuts a [start, end) range into a new MP4 and uses no provider inference, only worker ffmpeg. Video filter runs a pixel pass: ops[] such as dim or crop, or an allowlisted filtergraph. Neither generates new picture content.

The YouTube page lists beauty filters, color adjustment and audio repair as its examples. It does not mention trimming, cropping or dimming by name, so treat those as consistent with the minor-edit idea and not as explicitly listed. They do not make a person say or do anything and do not invent a scene.

When a regeneration changes the answer

Once a tool generates pixels that were not in the source, check the four triggers. A regenerated background that looks like a real place, a face that now says new words, or an inserted object in real footage all move the clip toward Yes. A stylized, obviously unreal result does not.

• Did the edit only adjust what the camera captured? Likely minor.
• Did it add a realistic thing that was never there? Check the triggers.
• Is the real event or place now different from what happened? Disclose.
• Is the result fantastical or fully animated? The page says no disclosure.

A worked example

Take a 30-second product video. You trim it to 24 seconds, dim the intro and correct the color: all of that is plain ffmpeg work on the camera footage, and the exempt examples on the page point the same way. Now regenerate the background as a realistic street that does not exist. That is a realistic scene that did not occur, so the answer is Yes.

The same file can have both kinds of edit. Judge the strongest edit in the file, not the average one.

When unsure

The page says YouTube may auto-apply labels as well, so over-disclosing a borderline clip is cheap given that disclosure does not limit reach or earnings.