AI influencer licensing: how to license your likeness to brands

AI influencer licensing means a creator gives a brand permission to use their face, and sometimes their voice, in AI-generated content, in exchange for a fee. The licence sets where the content can run, for how long, and what it may never show. The safest licences add one more rule: every finished asset needs the creator’s approval before it is published.

This guide is for creators deciding whether to license their likeness, and for brands that want to do it properly. It is general information, not legal advice.

What “AI influencer” means here

The phrase is used for two different things:

• Virtual influencers: characters that don’t exist, built entirely with CGI or AI.
• Real creators whose likeness is used by AI: a brand generates new images or video of a real person, with their permission, instead of booking a shoot.

This guide is about the second. When the person is real and recognisable, consent, scope and disclosure stop being nice-to-haves. They are what make the content lawful.

Three ways brands license a likeness today

Model	How it works	Creator control	Main risk
Buyout or model release	One payment for a broad grant, often to an “AI actor” library	Low: signed once, used anywhere	Your face appears in ads you never saw, for products you’d refuse
Digital twin access	The brand gets a trained model of you and generates freely within the contract	Medium: limited by contract, not by the workflow	Hundreds of variations, no way to check each one
Per-asset approval	The brand generates; you approve each finished file before it runs	High: nothing runs without a yes	Slower for brands that iterate many variants

The difference that matters is when your consent is given. A release or a twin contract is consent up front, to things nobody has seen yet. Per-asset approval is consent to the exact file that will be published.

What a good likeness licence defines

Whatever platform or lawyer you use, a likeness licence should answer these questions in writing:

1. Channels. Paid social, organic social, display, out-of-home, TV, print. Name them; “all media” is too broad.
2. Territories. Countries or regions where the content may run.
3. Term. A start and end date. Content outside the term must come down.
4. Categories. Which product categories are allowed. Exclusions are just as important: alcohol, gambling, politics, competitors.
5. Hard no’s. Anything you won’t be shown doing, wearing or saying, regardless of category.
6. Approval. Whether each asset needs your sign-off, how long you have to answer, and what happens if you don’t. Silence should never count as consent.
7. Labelling. How the content is disclosed as AI-generated, visibly and in its metadata.
8. Training and reuse. Whether your images may train or fine-tune models, and what happens to them when the licence ends.
9. Revocation. How you can end the licence early, and what happens to content already published.
10. Payment. The fee, when it’s paid, and whether it’s held in escrow until you approve.

LikePact publishes a free likeness licence template that covers each of these points.

Clauses to refuse or renegotiate

• “In perpetuity” or “throughout the universe.” Ask for a fixed term and named territories.
• “All media now known or hereafter devised.” This covers formats that don’t exist yet. Name the channels.
• Training rights by default. Permission to make an ad is not permission to train a model on your face. Ask for this to be excluded, or priced separately.
• Sublicensing. The brand should not be able to pass your likeness to its agencies, partners or AI vendors without your consent.
• No approval step. Without approval, your only protection is enforcing the contract after the content is already out.
• Vague descriptions of use. In California, a contract provision allowing a digital replica of a performer can be unenforceable unless it describes the intended uses reasonably specifically (AB 2602). Specific is safer for both sides everywhere.

How to price your likeness

There is no standard rate card yet. These factors usually move the price:

• Reach and audience fit. The value of your face comes from who already knows and trusts it.
• Channel. Paid media and out-of-home are usually worth more than organic posts.
• Exclusivity. If the licence stops you working with competitors, charge for it.
• Term and territory. Longer and wider means more.
• Volume. Price per approved asset, or per licence with a cap on the number of assets.

A simple way to start: set a minimum fee per licence, a cap on the number of assets, and a price per extra asset. Review it after your first few deals. On LikePact, creators set a minimum fee in their standing terms and keep 85% of each licence fee.

The law in brief

Rules differ by country, and they are changing fast. In outline:

• EU, data protection. An image of an identifiable person is personal data under the GDPR. Using it needs a lawful basis, usually consent, and the person can withdraw that consent.
• EU, AI Act. AI-generated content showing a real person in a way that appears authentic is a “deep fake” under the EU AI Act. From 2 August 2026, whoever publishes it must disclose that it is AI-generated (Article 50(4)). Our Article 50 guide for advertisers covers the details.
• Sweden. The Names and Pictures in Advertising Act (1978:800) already requires a person’s consent before their image is used in marketing.
• United States. There is no single federal likeness law yet. States protect it through the right of publicity. Tennessee’s ELVIS Act covers voice and likeness. California and New York limit contracts for digital replicas that don’t describe their uses clearly. Federal proposals, such as the NO FAKES Act, are still pending.
• Platforms. YouTube, TikTok and Meta all require realistic AI-generated content to be disclosed with their own tools, in addition to any legal duty.

How LikePact handles it

LikePact is built around per-asset approval:

1. Creators set standing terms: channels, territories, excluded categories, a minimum fee, a maximum term and their hard no’s.
2. Brands request a licence inside those terms, and the creator accepts or declines it.
3. Every asset is checked against the licence before the creator sees it, then sent for approval. Unanswered requests expire after 72 hours. They never auto-approve.
4. On approval, the asset is labelled as AI-generated, signed with a C2PA manifest, and recorded in an append-only consent ledger. Then the creator’s fee is released.

Frequently asked questions

Is an AI likeness licence the same as a model release?

No. A model release is usually a one-off grant covering a shoot. An AI likeness licence covers new content that doesn’t exist yet, so it needs limits on channels, territory, term and categories, and ideally approval of each asset.

Can I license my face without creating a digital twin?

Yes. With per-asset approval, brands use their own tools and you approve each result. You don’t hand over a model of yourself that can be used freely.

Can a brand train an AI model on my photos?

Only if your licence allows it. Many contracts include training rights by default, so read for words like “train”, “fine-tune”, “improve our models” and “derivative works”.

What happens if I want to stop?

That depends on your licence. On LikePact, you can revoke a licence with 30 days’ notice. Assets you already approved stay covered for the rest of their term, and nothing new can be approved.