You cannot un-publish. The cheapest place to catch an accidental likeness or a copyrighted character is before a video ships — not after a complaint, a takedown, or a lawyer's letter. Here is a repeatable pass you can run, or wire into your pipeline, so screening is a gate rather than a fire drill.
Scope it first
The pass applies to anything you publish or monetise — clips you generate, and user-generated video you host or amplify. If your name ends up on it, it is in scope. “A user uploaded it” is a weak position once it is on your surface.
The checklist
- Sample the whole video, not just the hero shots. Accidental likeness and stray IP hide in backgrounds and brief appearances — exactly where a human skim misses them.
- Screen for public-figure resemblance. Faces that match a real person are the headline risk — and a “fictional” face is not always fictional.
- Screen for copyrighted characters too. Not just faces — a recognisable cartoon, game hero, or mascot is an IP exposure of its own.
- Treat every match as a candidate for review, never a verdict. The tooling is probabilistic and over-eager to name; a human confirms, with the frame in front of them.
- Rank by confidence and work the top. A face-verified match is a finding; an unnamed detection is a lead. Read the list in order and you triage the real risk first.
- Clear your knowns once. Your staff, your licensed talent — mark them so a re-run doesn't re-flag the same cleared person every time.
- Keep the receipt. A timestamped record of what you screened, and what you found, is itself part of a diligence defence.
When something flags
Don't panic-delete. Assess context: would a reasonable audience read this as a real person, or as coincidence? The more corroborating signals — profession, setting, storyline — the more a resemblance reads as a reference. That is the “reasonable person” test; for anything that clears it, talk to counsel before you ship.
Make it a gate, not a heroic effort
The point of a checklist is that it runs every time, the same way, without depending on whoever happened to watch the cut. Put the screen in the pipeline, ahead of publish — and keep the record it produces.