Face swap for ads: consent, disclosure and review checklist

This is a production checklist, not jurisdiction-specific legal advice. Laws, contracts and ad-platform rules can differ by location and campaign, so have the responsible legal or policy owner approve the actual use.

Face-swap ad go/no-go check

• Go: the person and source-footage owner have approved the exact commercial use; the ad does not invent an endorsement or product experience; claims are substantiated; required disclosure is present; and the final render passes human review.

• No-go: the likeness came from the open web, a stock license is silent about synthetic identity edits, the person appears to say or demonstrate something they did not approve, or ownership and platform treatment are unclear.

1. Build the rights record before production

Store the agreement or approval with the creative brief. A useful rights record identifies:

• Identity: whose face, voice and performance are used.

• Source assets: who owns the base video, face image, music, logos and other media.

• Permitted use: campaign, product, claim, format and allowed synthetic edits.

• Distribution: channels, paid or organic placement, territories and dates.

• Operations: who may generate, review, download, retain and reuse the assets.

• End of use: what happens to live ads, derivatives and stored inputs when the permission expires or is withdrawn under the agreement.

Do not assume a general appearance release, influencer agreement or stock license covers a newly generated performance. Read the actual language and obtain a specific amendment when the scope is unclear.

2. Keep endorsements and claims truthful

A face swap can change the net impression of an ad even when the spoken words stay the same. The FTC explains that a likeness can form part of an endorsement and that an AI stock avatar is not automatically prohibited, but a celebrity avatar presented as a favorable testimonial without permission can violate the rule. Its current reviews and testimonials guidance also warns that an actor portraying a testimonialist can still be deceptive depending on the presentation.

• Approve the exact script and action. Do not make a real person appear to use, recommend or experience a product unless that representation is authorized and truthful.

• Substantiate every claim. Keep the same evidence standard for a synthetic performance as for a filmed one.

• Identify dramatization when needed. A disclosure should clarify the impression a reasonable viewer could otherwise misunderstand; it does not cure a false claim or unauthorized endorsement.

Google Ads gives AI-generated video of a popular creator falsely endorsing a get-rich-quick offer as an unacceptable-business-practices example. Its broader misrepresentation policy prohibits manipulated media used to deceive, defraud or mislead.

3. Decide disclosure for each destination

Do not copy one disclosure rule across every platform. Record the campaign destination, policy URL, check date, selected setting and visible label in the launch ticket.

• TikTok ads: TikTok’s current Ads Manager guidance says its AI-generated-content disclaimer is mandatory for AI-generated, synthetic or significantly AI-modified image, video or audio.

• Google Ads: general misrepresentation rules apply, while election ads have additional synthetic-media disclosure requirements. Review the policy that matches the actual campaign type.

• Other channels: check the current ad, branded-content, synthetic-media and political-content policies for the exact placement before upload.

TikTok’s current ad-disclaimer instructions explain where to enable the label. Recheck live requirements at launch because platform policies change.

4. Generate a controlled pilot

• Choose a representative clip. Include speech or expression, head movement, partial profile and at least one difficult lighting change.

• Use approved inputs. Keep the exact asset versions tied to the rights record.

• Generate one short draft. In Magic Hour’s video face-swap tool, upload the base video and the approved face image, then render the smallest clip that tests the real creative.

• Automate only after approval. If the pilot passes and volume justifies integration, use the face-swap API documentation and retain the same approval gates.

5. Review the final output, not only the source frames

• Identity: the face is the approved person and does not drift toward someone else.

• Frame quality: inspect eyes, teeth, jaw, hairline, occlusion, fast motion, profile views and scene cuts.

• Performance: expression and lip movement do not alter the intended meaning or tone.

• Claims: spoken, written and implied claims match the approved brief and evidence.

• Context: the person is not placed in a location, action, relationship or endorsement they did not approve.

• Disclosure: the selected platform setting and any visible label survive cropping, captions and placement previews.

• Landing page: offer, brand, product and claims match the ad.

6. Preserve an auditable launch record

Keep the approved script, source-asset IDs, rights record, generated file, reviewer names, platform-policy links and check dates, disclosure decision, final placement preview and campaign dates together. If a later edit changes the person, claim, product, market, media buy or context, run the approval again.

When face swap is the wrong production choice

• Real testimonial: film the actual customer or spokesperson when credibility depends on a real experience.

• Unclear rights: use a fully licensed actor, approved stock avatar or original non-identifiable character instead.

• Sensitive or consequential claim: use direct evidence and a real authorized expert or spokesperson where the context demands it.

• Poor pilot quality: reshoot or use another creative concept rather than publishing an uncanny or misleading result.

Continue learning

Continue learning: best AI face swap tools, best free face swap tools, and step-by-step face swap workflow.