AI likeness law, answered

Is it legal to use AI-generated models in advertising?

Generally yes. Neither US federal law nor EU law bans AI-generated people in advertising as such. What decides legality is everything around the image: whose likeness it uses and whether they consented, whether the ad must disclose that a person is AI-generated, and whether the ad is truthful. Ordinary advertising law still applies, so an AI model cannot be used to show a result the product does not deliver, or to imply that a real person endorses something they have not. Rules differ by country and by US state, and several took effect in 2025 and 2026, so check each market where the ad will run.

On Mirai: Every likeness on Mirai belongs to a named person who has consented, and each licence records the format, term and territory it covers.

What is the legal risk of using a fully synthetic AI model?

The main risk is that nobody can sign a release. A synthetic face has no owner to grant rights or confirm consent, so a brand cannot show a chain of permission if the image is challenged. The face may also resemble a real person, a right of publicity risk you cannot fully rule out in advance. Copyright in the image may be thin, and New York now requires disclosure of synthetic performers in some ads. If a synthetic person is presented as a customer giving a testimonial, the FTC's 2024 rule on consumer reviews and testimonials, which covers testimonials from people who do not exist, is also relevant.

On Mirai: Mirai works only with real people who have signed a likeness agreement, so the permission exists before the image does.

Can an AI-generated face accidentally look like a real person?

Yes. Image models learn from very large sets of real photographs and can produce faces that closely resemble real people, including well-known ones, without anyone intending it. Whether that creates liability depends on the jurisdiction, but in the US the usual questions are whether a real person is identifiable and whether the ad trades on their identity. Courts found liability for lookalikes long before generative AI, for example in the New York case Onassis v. Christian Dior in 1984. There is no practical way to check a synthetic face against every person alive, so the risk can be reduced but not eliminated.

On Mirai: Every Mirai likeness is of a named person who signed for its use, so there is no unknown resemblance to clear.

Do you need consent to create an AI likeness of a real person?

For commercial use, almost always. Most US states protect a person's name and likeness, and often their voice, against unauthorised commercial use, through statutes such as New York Civil Rights Law sections 50 and 51, California Civil Code section 3344 and the Illinois Right of Publicity Act, or through common law. Federal false endorsement claims are also available under the Lanham Act. The UK has no standalone image right, but passing off and data protection law can apply, and many EU countries protect a person's image directly. Consent to a photoshoot does not automatically cover AI use: newer laws expect the digital use to be described specifically.

On Mirai: Every person on Mirai signs the Model Agreement before their AI likeness is created, and sets the categories, brands and territories it cannot be used for.

What does the New York Fashion Workers Act require for AI replicas of models?

The Fashion Workers Act, in force since 19 June 2025, requires clear, conspicuous and separate written consent before a model management company or a client creates or uses a model's digital replica. The consent must set out the scope and purpose of the use, the rate of pay and how long the replica will be used. A digital replica is a significant computer-generated or AI-enhanced representation of a model's likeness that substantially replicates or replaces their appearance or performance; routine retouching is not covered. Whether it applies depends on the engagement, not only on where the model lives.

On Mirai: Mirai has a New York Addendum for engagements the Act covers, and each licence records its format, term and territory against the talent's signed agreement.

What do California's AB 2602 and AB 1836 change?

Both took effect on 1 January 2025. AB 2602 targets vague AI clauses in performer contracts: a term that lets a digital replica of a person's voice or likeness do work they would otherwise have performed in person is unenforceable where the intended uses are not reasonably specifically described and the person was not represented by a lawyer or a union in the negotiation. AB 1836 extends California's right of publicity for deceased personalities to digital replicas, so using a replica of a deceased person in an audiovisual work or sound recording generally needs the estate's consent, subject to exceptions such as news, commentary, satire and parody.

On Mirai: Mirai licences name the specific format, term and territory of each use rather than relying on a blanket grant.

Which other US states regulate AI likeness?

Several, and the list is growing. Tennessee's ELVIS Act, in effect since 1 July 2024, added voice to the state's right of publicity and created liability both for publishing a person's voice or likeness without authorisation and for distributing a tool whose primary purpose is producing a specific person's likeness or voice without authorisation. Illinois requires written consent for commercial use of a person's identity under its Right of Publicity Act, which was amended in 2024 to address digital replicas. Illinois's Biometric Information Privacy Act separately requires written consent before collecting biometric identifiers such as face geometry; how it applies to generative likeness tools is not yet settled. Other states have narrower laws, often on election deepfakes or intimate imagery.

Is there a federal AI likeness law in the US?

Not a general one. The NO FAKES Act, which would create a federal right to control digital replicas of a person's voice and visual likeness along with a notice-and-takedown system for platforms, has been introduced in Congress but had not become law when this page was last reviewed. The TAKE IT DOWN Act, signed in May 2025, is narrower: it makes it a federal crime to publish non-consensual intimate images, including AI-generated ones, and requires covered platforms to remove them on request. For everything else, brands rely on state right of publicity law, the Lanham Act for false endorsement and the FTC's rules against deceptive advertising.

Do you have to label AI-generated people in ads in New York?

In some cases, yes. A New York law enacted in late 2025, and in effect from 2026, requires anyone producing an ad who knows it includes a synthetic performer to disclose that conspicuously. As written, a synthetic performer is a digitally created asset meant to look like a human performer who is not recognisable as any identifiable real person, so the rule is aimed at invented people. An AI likeness of a named, consenting person sits differently, but that point has not been tested, and other duties may still apply, including the EU AI Act for ads shown in the EU and the FTC's rules on misleading advertising. Disclosing is often the lower-risk choice.

What does the EU AI Act require for AI-generated images?

Article 50 of the EU AI Act sets transparency duties that apply from 2 August 2026. Providers of AI systems that generate images, video or audio must mark the output in a machine-readable way so it can be detected as AI-generated. Deployers who publish a deep fake, meaning AI content that resembles real people, objects or places and would falsely appear authentic, must disclose that it was artificially generated or manipulated. For evidently artistic, creative or fictional work the disclosure can be lighter, as long as it does not spoil the work. Consent from the person shown does not remove the duty. Check transitional rules and Commission guidance for the system you use.

On Mirai: Every Mirai licence has a public verification page, and embedded Content Credentials (C2PA) that mark licensed files as AI-generated in a machine-readable way are being rolled out.

Are licensed digital twins the same as deepfakes?

Not in the sense most people mean. A deepfake usually means a realistic fake of a real person made without their permission, often to deceive. A licensed digital twin is made with the person's written consent, used within limits they agreed and paid for. Legally, though, the labels can overlap. The EU AI Act defines a deep fake by whether content resembles a real person and could falsely appear authentic, not by consent, so a licensed likeness can still need a disclosure. Consent answers the likeness question and disclosure answers the transparency question. Most commercial uses need both.

On Mirai: Every Mirai licence has a public verification page showing that the likeness was licensed and the scope of the licence.

Who owns the copyright in an AI-generated image?

In the US, possibly no one. The US Copyright Office's January 2025 report on copyrightability concluded that material generated by AI without sufficient human control is not protected, and that prompts alone are generally not enough. In March 2025 a federal appeals court confirmed in Thaler v. Perlmutter that copyright requires a human author. Human contributions can still be protected, such as your own photographs used as inputs, creative selection and arrangement, or substantial edits. Other countries differ; the UK, for example, has a specific rule for computer-generated works. Because copyright may be thin, a brand's strongest protection often comes from its licence and from the likeness rights of the person shown.

On Mirai: The licence sets what the brand may do, and the person shown keeps their likeness rights, which do not depend on copyright.

Can fans make AI images of copyrighted characters?

Not without legal risk unless the owner allows it. Well-defined characters can be protected by copyright, and their names and designs are often registered trade marks. Making and sharing new images of a character is generally a use that needs permission, unless an exception applies, such as fair use in the US or the parody, caricature and pastiche exception in the UK. Those exceptions depend on the facts, and commercial use weighs against them. Many owners tolerate non-commercial fan art, but tolerance is not a licence and can be withdrawn. Whether training AI on characters is lawful is a separate question that courts are still deciding.

On Mirai: In Character Studio, IP owners set the references and rules for their characters, fans create within those rules, and commercial use needs a separate licence.

Are public-domain characters free to use commercially?

The public-domain version is, but check what else still protects it. Copyright expires at different times in different countries, so a character can be in the public domain in the US and still protected elsewhere. In the US, works published in 1928, including the first Mickey Mouse film, entered the public domain on 1 January 2024, but later versions of a character with new traits stay protected until their own copyright expires. Trade mark law can also still apply, so using a character in a way that suggests the owner endorses your product can be a problem even after copyright ends.

On Mirai: Mirai records the copyright basis for each public-domain character licence on that licence's public verification page.

How does Mirai record consent and licence scope?

Every person on the roster signs Mirai's Model Agreement before their AI likeness is created, and the version they signed is recorded. Talent set restrictions on categories, brands and territories. Restricted categories and brands are screened before anything is generated, and territory conflicts are checked when a licence is bought. Each licence records the format, term and territory it covers, and print and out-of-home follow the talent's own setting: on demand inside her rules, or her individual approval. Talent are paid for each licence. A model can withdraw consent to stop new generation on 72 hours' notice, and licences already granted continue for their agreed term.

Why are Mirai images watermarked until licensed, and what do Content Credentials show?

Because a generated draft is not a licence. Images and videos made on Mirai stay visibly watermarked while they are drafts, so an unlicensed file cannot pass as cleared work. Once a licence is bought and any required approval is given, the brand downloads the clean file. Every licence has a public page anyone can check, and embedded Content Credentials (C2PA), a manifest in the file that marks it as AI-generated and links to that page, are being rolled out. Social platforms often strip embedded metadata, so the public verification page is the durable record. Content Credentials show provenance; on their own they do not satisfy every disclosure rule.