How to Use Ai‑generated Celebrity Likenesses for Advertising in China (2026)

AI celebrity likeness China compliance has become one of the most pressing questions for advertisers, platforms and talent agencies in 2026, as deep synthesis management rules, the generative AI service measures and AI‑content labelling requirements, combined with heightened personality‑rights enforcement, reshape what brands can lawfully do with synthetic media. Whether you are commissioning a fully AI‑generated endorsement, cloning a voice, or reanimating archival footage, the legal exposure now spans civil personality rights, advertising truthfulness obligations and mandatory content‑labelling duties. This guide sets out a practical, step‑by‑step clearance workflow, with required documents, timelines, costs and a risk checklist, so compliance teams can move from concept to launch without triggering infringement claims or regulatory penalties.

It is written for general counsel, marketing leads and non‑lawyer compliance staff who need actionable procedure, not abstract theory.

Disclaimer: This article is general information and last reviewed on 14 September 2026. It is not a substitute for bespoke legal advice on any specific campaign.

1. Eligibility, when you can (and cannot) use an AI celebrity likeness

The threshold question for any AI celebrity likeness China project is whether you have the legal right to depict a recognisable individual at all. In the People’s Republic of China, an individual’s portrait and personality are protected as civil rights under the Civil Code, and the commercial use of a recognisable likeness, including one generated or altered by AI, engages those rights regardless of the technology used to produce the image. The fact that no camera photographed the person does not remove the protection: what matters is whether the output is identifiable as that person.

When consent is required

Consent is required whenever you create, reproduce or commercially exploit a likeness that the public would recognise as a specific celebrity. This includes AI‑generated still images, video, synthetic voice performances and composite works that combine several traits into an identifiable persona. For advertising specifically, consent is effectively non‑negotiable: commercial endorsement is the paradigm case of unauthorised exploitation that PRC courts have repeatedly restrained and compensated. Assume consent is required for every advertising use unless a narrow statutory exception plainly applies.

Exceptions and editorial uses

Limited exceptions to portrait‑right protection exist under the Civil Code for uses such as genuine news reporting, public‑interest purposes, or scholarly analysis. These editorial carve‑outs are narrow and do not extend to advertising. A campaign that dresses commercial promotion as “editorial” content will not qualify. Where an AI likeness is satirical or transformative, some latitude may exist, but any use tied to selling a product or service falls squarely within the consent requirement. In practice, if the objective is to move product, treat consent as mandatory.

2. Legal bases: personality rights, copyright, portrait rights and advertising law

Understanding the AI celebrity likeness China legal framework requires mapping several overlapping regimes. No single statute governs synthetic endorsements; instead, civil personality rights, copyright, and advertising regulation operate together, supplemented by sector‑specific rules on synthetic media.

Civil Code, personality and image rights

The Civil Code of the People’s Republic of China (in force since 1 January 2021) establishes personality rights, including the right to one’s portrait, name, voice and reputation. Unauthorised use of a person’s portrait for commercial gain infringes these rights and exposes the user to civil remedies including injunctions, cessation of infringement, apology and damages. The Code protects the portrait right expressly, and its personality‑rights provisions extend protection of the voice by reference to the rules protecting portraits, a critical point where voice cloning is involved. Because the right attaches to identifiability rather than to any particular photograph, AI‑generated depictions and AI‑altered footage both fall within its scope.

Advertising Law obligations and false or misleading claims

The Advertising Law of the People’s Republic of China, enforced by the State Administration for Market Regulation (SAMR), imposes independent obligations. Advertisements must be truthful and must not mislead consumers. An AI‑generated endorsement that implies a celebrity personally uses or approves a product, when they do not, risks being treated as a false or misleading advertisement, quite apart from any personality‑rights claim. Endorsers themselves carry statutory responsibilities under the Advertising Law, so misrepresenting endorsement through synthetic media creates layered exposure. Advertisers should ensure that any depicted “endorsement” is genuinely authorised and accurately reflects the underlying commercial relationship.

Copyright adds a further layer. Where AI‑altered archival footage is used, the underlying film or photograph may itself be a copyrighted work owned by a studio, broadcaster or photographer. Clearing personality rights does not clear copyright in the source material, and both must be addressed. The China National Intellectual Property Administration (CNIPA) administers patents and trade marks, while copyright administration in China sits with the National Copyright Administration; both may be relevant to the IP‑layer issues in a synthetic‑media campaign.

3. Consent: what you must get and how to draft it for ai celebrity likeness china campaigns

Consent is the single most important document in any AI celebrity likeness China campaign, and generic celebrity licences drafted for conventional photo shoots are inadequate. They rarely contemplate synthetic generation, model training, or the persistence of a digital likeness. You need an AI‑specific licence.

Minimum consent wording

A robust AI likeness licence should expressly address each of the following:

• Express AI/synthetic permission. Explicit authorisation to generate, synthesise, and animate the likeness using AI, not merely to “use” existing images.
• Scope of use. Defined media channels, territories and campaign duration, with anything outside the defined scope excluded.
• Reputation protections. Approval rights over final creative so the depiction does not damage the person’s reputation (note that PRC personality rights are generally not freely waivable in the way some other systems permit).
• Sublicensing rights. Whether the brand may sublicense to agencies, media buyers or platforms, and on what terms.
• Compensation. Fees, royalties and payment triggers, documented to evidence consideration.
• AI model training and data retention. Whether the celebrity’s images or voice may be used to train or fine‑tune a model, how long training data is retained, and deletion obligations at term end.
• Vendor warranties and disclosure obligations. Requirements that the AI vendor warrant training‑data provenance and that outputs carry required synthetic‑media labelling.
• Indemnities. Allocation of liability if the likeness use gives rise to third‑party claims.

Talent and agent signoff

Confirm that whoever signs actually controls the rights. A talent agent can grant a licence only within the scope of their agency mandate, so verify the agency appointment and any exclusivity arrangements before relying on an agent’s signature. Where the celebrity signs personally, authenticate the signature and retain identity evidence. For high‑value deals, obtain both the agent’s execution and the principal’s confirmation to close off any authority gap. Do not rely on informal email approvals for commercial exploitation.

4. Step‑by‑step clearance workflow for ai celebrity likeness china

The following workflow is the operational core of any AI celebrity likeness China compliance programme. Each step has an owner and a documented output, so approvals are traceable and defensible if later challenged.

1. Scope and risk assessment (internal). Brand legal, marketing and compliance jointly identify the celebrity, the use case, the media channels, target territories, risk tolerance and budget. Output: a written scoping memo.
2. Confirm the rights chain. Clearance counsel verifies that the persona is controlled by the party granting rights, checks agency contracts, and screens for existing endorsements or exclusivity clauses that could conflict.
3. Consent negotiation and contract execution. In‑house counsel, external entertainment counsel and the talent agent negotiate the AI‑specific clauses above, execute with authenticated signatures, and collect supporting identity and authority evidence.
4. IP and data audit of the AI model. The technology vendor and external counsel confirm training‑data provenance, in particular whether the model was trained on the celebrity’s images or audio, and secure written vendor warranties.
5. Regulatory check and disclosure planning. Compliance counsel determines whether the content requires labelling under the deep synthesis and AI‑content labelling rules, what platform disclosure requirements apply, and whether any prior filing or security assessment is needed.
6. Production and metadata controls. The creative team and platform engineers embed watermarking and metadata, and preserve source files and audit logs.
7. Pre‑launch legal review and signoff. Legal, compliance, PR and the talent agent complete a final sign‑off checklist with documented approvals.
8. Monitoring and takedown plan. Platform operations and legal establish monitoring, takedown procedures and a crisis‑response protocol before launch.

Assign each step a named owner and a target date. A single unassigned step is the most common cause of a stalled or non‑compliant launch.

Teams should maintain a downloadable risk checklist mirroring these steps, with each item ticked off and dated before launch approval is granted.

5. Required documents and evidence

Documentation is not administrative overhead, it is the evidentiary backbone if a personality‑rights or advertising claim later arises. Retain originals and version histories. The following table sets out the core file set for an AI celebrity likeness China campaign.

Digital evidence and metadata provenance

Preserve the provenance of every asset. Store cryptographic hashes and timestamps of source files, generated outputs and disclosure metadata. If a dispute arises, contemporaneous logs demonstrating what was authorised, when, and by whom are far more persuasive than reconstructed accounts.

Sub‑licence chain

Where rights pass through agencies, media buyers or platforms, document the full sub‑licence chain. A break in the chain, for example, a platform running the creative beyond the licensed territory, can expose the brand even where its own conduct was compliant.

6. Timeline, deadlines and filings for foreign celebrities

A modest domestic campaign can typically clear in one to two weeks. High‑profile talent deals or cross‑border uses often take two to six weeks once negotiation, vendor audits and regulatory checks are factored in. Build in contingency: a single failed rights‑chain check or a vendor unable to warrant training‑data provenance can add days or weeks.

Where the celebrity is foreign, additional steps apply. You must consider the governing law and jurisdiction of the licence, foreign publicity or personality‑rights protections in the individual’s home country, any restrictions on cross‑border transfer of personal data used to build or train the likeness (which engage China’s Personal Information Protection Law and related cross‑border transfer rules), and platform‑specific requirements for synthetic content. Cross‑border deals frequently require a separate clearance track and should not be compressed into a domestic timeline. Establish platform takedown response times in advance so that, if unauthorised copies proliferate, removal can be actioned quickly rather than negotiated from scratch under pressure.

7. Costs and fees

Budget depends heavily on the celebrity’s tier, the media channels, the territory and the duration. The ranges below are indicative planning considerations rather than quotations.

Treat legal and compliance spend as risk‑reduction, not overhead: the cost of clearance is materially lower than the combined exposure of civil damages, administrative penalties and reputational harm from an unauthorised launch.

8. What changed in 2026, key regulatory updates you must know

The defining shift for AI celebrity likeness China compliance in 2026 is the maturing synthetic‑media regime overseen by the Cyberspace Administration of China (CAC). The framework builds on the deep synthesis provisions, the generative AI service measures, and, significantly, the mandatory labelling measures for AI‑generated content, which took effect on 1 September 2025. These rules target the generation and dissemination of synthetic media, including AI‑generated images, video and voice, and impose obligations to label or disclose synthetic content used in public communications, both through visible markings and embedded metadata. Advertisers can no longer treat labelling as optional creative garnish; where content is synthetic, disclosure is a compliance requirement.

Platform operators carry parallel duties to identify and manage synthetic content on their services, which flows through to advertisers via platform disclosure and review requirements. The practical effect, industry observers expect, is that platforms will increasingly refuse or pull synthetic endorsements that lack proper labelling and documented consent. Enforcement attention has also sharpened on personality‑rights infringement and misleading endorsements, meaning the two exposure vectors, civil claims and regulatory penalties, now tend to arrive together. For advertisers, the message is straightforward: labelling, consent documentation and provenance controls must be built into the campaign from the outset, not retrofitted after a complaint.

9. Common pitfalls and risk mitigation checklist

The recurring failures in AI celebrity likeness China campaigns are predictable and preventable:

• Relying on a conventional photo licence. Standard celebrity agreements rarely cover AI generation or model training, always use an AI‑specific licence.
• Assuming the agent has authority. Verify the agency mandate and exclusivity before relying on any signature.
• Ignoring copyright in archival footage. Clearing personality rights does not clear the underlying film or photograph.
• Skipping vendor provenance warranties. Without warranties, audit rights and indemnities, you inherit the vendor’s training‑data risk.
• Omitting synthetic‑media labelling. Failure to disclose can breach the deep synthesis and AI‑labelling rules and trigger platform takedowns.
• Overrunning territory or term. Media buys that exceed the licensed scope create fresh infringement even where the base licence was valid.
• Neglecting voice consent. Voice cloning requires its own release; a portrait licence does not cover it.
• Treating “editorial” as a shield. Advertising does not qualify for editorial exceptions.
• Failing to retain evidence. Without dated logs and originals, defending a claim becomes far harder.
• No monitoring or takedown plan. Unauthorised copies spread quickly; establish response procedures before launch.

Mitigation rests on three pillars: watertight consent documentation, vendor warranties backed by audit rights and indemnities, and an escalation matrix that names who decides on takedowns, corrective disclosures and crisis communications. Consider media liability insurance for high‑risk campaigns. Where disputes escalate, structured dispute resolution, including options familiar from art arbitration in China and sports arbitration in China, may offer a route to resolve personality‑rights and endorsement disputes efficiently, subject to the arbitrability of the particular claim.

10. Comparison: real‑photo consent vs AI‑generated likeness vs AI‑altered archival footage

Different production methods carry different legal and practical risk profiles. The table below highlights the distinctions that should drive your contract addenda.

Conclusion and next steps

Getting AI celebrity likeness China compliance right in 2026 comes down to disciplined process: confirm the rights chain, secure an AI‑specific consent and licence, audit the model’s training‑data provenance, plan mandatory synthetic‑media disclosure, and preserve evidence throughout. The regulatory environment now couples civil personality‑rights exposure with active advertising and synthetic‑media enforcement, so shortcuts that once passed unnoticed are increasingly likely to surface as claims or takedowns. Advertisers, platforms and talent agencies that build consent documentation, vendor warranties and labelling controls into the workflow from day one will move faster and safer than those retrofitting compliance after a complaint.

For bespoke clause drafting, cross‑border clearance and enforcement strategy, request a match with a technology and IP specialist through the GLE lawyer directory.

Sources

1. National People’s Congress (NPC)
2. Cyberspace Administration of China (CAC)
3. State Administration for Market Regulation (SAMR)
4. China National Intellectual Property Administration (CNIPA)
5. Supreme People’s Court (SPC)
6. China Judgments Online