Key Takeaways
- A virtual-influencer likeness license must clearly define scope, term, AI training rights, derivative works, compensation, disclosure duties, and termination rights. Clear drafting prevents silent expansion of usage.
- Brands face right-of-publicity, copyright, and disclosure risks in 2026. Independent creation documentation and real-time monitoring now function as baseline safeguards.
- Usage rights, training rights, and derivative rights require separate clauses and separate fees. Courts and creators treat silence on AI training as a prohibition.
- Perpetual grants, vague moral-rights language, and failure to document human authorship are common contract pitfalls that drive disputes and reputational damage.
- Sozee’s locked-likeness platform enforces consistency, disclosure, and data isolation at the asset level, which reduces licensing disputes. Explore Sozee’s compliance-first platform.
Seven Core Terms for a Virtual-Influencer License
A compliant virtual-influencer likeness agreement rests on seven foundational clauses. Each clause addresses a specific risk surface identified by practitioners across Diverge Legal, Foley & Lardner, and the SAG-AFTRA Digital Replica Rider.
- Scope of Grant. Enumerate every permitted use, including platform, media type, territory, and campaign name. Treat rights not listed as reserved. Broad catch-alls such as “in any media now known or hereafter devised” combined with “perpetual” and “irrevocable” are red-flag language that quietly expand AI training rights.
- Term and Renewal. Time-limit every grant and specify renewal options. Perpetual grants are increasingly viewed as unenforceable or reputationally risky, especially as AI capabilities and use cases expand.
- AI Training Rights. State clearly whether the licensed assets may be used to train, fine-tune, or validate any AI model, internal or third-party. A contract silent on AI training rights is not neutral; courts and creators increasingly treat silence as a “no.”
- Derivative Works. Define whether the licensee may create derivative characters, style variants, or voice clones. Require separate written consent for each new derivative category to keep control over character evolution.
- Compensation Structure. Separate AI-related fees from base licensing fees. Compensation for AI-related uses must be addressed separately from standard day rates. Track revenue generated through AI-generated content and reserve audit rights for the licensor.
- Disclosure Obligations. Require the licensee to comply with all applicable disclosure mandates, including the FTC Endorsement Guides, New York S.8420-A, and platform-level AI labeling policies. Make compliance a condition that must be met before any asset is published.
- Termination and Reversion. Include a kill-switch mechanism that allows the licensor to revoke AI usage rights mid-term if usage exceeds the agreed scope. Define a wind-down period for assets already in distribution so both parties understand the exit path.
These seven clauses create the contractual foundation, but contracts alone cannot guarantee compliance. Three additional dimensions, real-person resemblance, copyright status, and disclosure requirements, determine whether a virtual-influencer campaign survives regulatory scrutiny.
Legal and Compliance Factors for 2026 Campaigns
Three compliance dimensions govern every virtual-influencer campaign in 2026.
Real-person resemblance. An AI character whose appearance is substantially similar to an identifiable living or deceased individual triggers right-of-publicity exposure. Brands must document that the virtual influencer is independently created and cleared, rather than an unlicensed imitation. They also need records of human creative decisions in persona development.
AI copyright status. U.S. copyright law requires human creative input for protection of AI-generated content under 17 U.S.C. § 102(a). Licensors must document the human authorship, including designers, art directors, and scriptwriters, that shaped the character to secure enforceable IP rights in the licensed persona.
Disclosure requirements. The FTC treats brands as the primary targets for disclosure violations, even when contracts delegate disclosure duties to agencies. FTC penalties reach up to $53,088 per violation, and class-action damages against major brands have been significant. Contracts set expectations, but brands also need active real-time monitoring within 24 hours of publication.
Breaking Down AI Usage, Training, and Derivative Rights
The table below distinguishes the three AI-rights categories that must be addressed separately in any virtual-influencer likeness agreement.
| Rights Category | What It Covers | Key Contractual Control |
|---|---|---|
| Usage Rights | Reproduction, distribution, and display of the licensed character across defined platforms, territories, and campaign types | Enumerate platforms and campaign names, and set a fixed term with no automatic renewal. Define platforms, duration, territories, media type, and whitelisting before creation. |
| Training Rights | Use of the character’s visual assets, voice samples, or behavioral data to train, fine-tune, or validate any AI or machine-learning model | Require prior written consent and a separate fee, limit to a defined Recording Period, and include a revocation mechanism plus data-deletion clause on termination. |
| Derivative Rights | Creation of new characters, style variants, voice clones, or other outputs derived from the licensed persona | Require separate written informed consent for each derivative category. Preserve moral rights including the right to object to uses that materially misrepresent the Performer Identity, and prohibit intimate or defamatory derivatives without explicit consent. |
Sample Clauses for Virtual-Influencer Likeness Licenses
The following clause language reflects current practitioner guidance and can serve as a starting-point template. Qualified counsel should review and adapt each clause before execution.
AI Training Prohibition (default position):
Synthetic Replica Prohibition:
Sublicensing and Third-Party Sharing:
Key negotiation points include excluding AI generation entirely as the default, requiring separate written consent and a separate fee for any AI-generated or altered version, and naming specific platforms where AI output may appear instead of accepting open-ended channel grants.
AI Training Rights in Influencer Contracts
AI training rights require dedicated provisions that go beyond standard licensing language. The following provisions reflect guidance from New York attorney Jillian Rhodes and the SAG-AFTRA Digital Replica Rider.
AI training provisions should address six connected controls, each closing a different gap in standard licensing language:
- Scope limitation. First, specify whether training rights extend only to internal brand models or also to third-party vendors and foundation model providers. AI training clauses must clarify this distinction explicitly because the economic and reputational stakes differ.
- Defined duration. Second, time-limit the grant. Perpetual training grants are increasingly viewed as unenforceable, and a defined Recording Period has become the recommended standard.
- Separate compensation. Third, separate AI training fees from ordinary licensing fees. Compensation for AI training rights must be separated from ordinary licensing fees, with additional payment required if the trained output is used commercially.
- Revocation mechanism. Fourth, include a contractual right for the licensor to withdraw AI training consent. This approach aligns with emerging statutory rights in multiple jurisdictions.
- Data deletion. Fifth, reserve the right to demand deletion of any recordings, images, or materials that could be used for AI training if the contractual relationship ends. This control supports both privacy and publicity-rights strategies.
- Audit rights. Sixth, include audit rights to allow verification of how licensed content is being used, including whether it has been incorporated into any AI training pipeline. Verification converts paper protections into enforceable practice.
Protect your AI training rights with Sozee’s locked-likeness workflow.
Right of Publicity for Virtual Characters
The right-of-publicity landscape for AI-generated personas has shifted materially in 2025–2026 across multiple jurisdictions.
At the state level, the following statutes directly affect virtual-influencer licensing:
- California SB 683, effective January 1, 2026, allows individuals to seek injunctions requiring removal of unauthorized AI-generated representations within two business days, with damages of the greater of $750 or actual damages plus attributable profits.
- California AB 1836 expands postmortem publicity rights to cover unauthorized digital replicas of deceased performers.
- New York S.8420-A, effective June 9, 2026, requires clear disclosure when advertisements include AI-generated synthetic performers and applies to any advertisement distributed to New York audiences, regardless of advertiser location.
- Tennessee’s ELVIS Act expressly extends publicity-rights protection to AI-generated voice replication and likenesses, with civil and some criminal penalties for unauthorized commercial use.
- Montana law addresses the use of names, voices, and visual likenesses.
At the federal level, the NO FAKES Act was unanimously advanced by the U.S. Senate Judiciary Committee on June 18, 2026. The bill would establish a licensable federal property right over voice and visual likeness in AI-generated digital replicas, create a DMCA-style notice-and-takedown process with $25,000 penalties for false counter-notifications, and preempt future state laws while preserving causes of action under state statutes existing as of January 2, 2025.
Common Contract and Compliance Pitfalls
The following errors appear repeatedly in virtual-influencer licensing disputes and contract audits.
- “We used a perpetual license — now what?” Perpetual grants on AI likeness rights are increasingly difficult to enforce and reputationally costly. The economic value of a digital twin grows with technological improvements, so a perpetual grant signed in 2023 may now cover uses the licensor never contemplated.
- “Our moral-rights clause was too vague.” Contracts that omit specific prohibitions on defamatory, sexually explicit, or politically sensitive uses of the AI character leave licensors exposed. Higher-profile talent contracts should include carve-outs restricting sensitive product categories and provisions addressing moral rights or reputational harm.
- “Our AI character looks too much like a real person.” A virtual influencer whose appearance closely resembles an identifiable individual triggers right-of-publicity claims even when no real person’s photos were used in training. Brands must document independent creation and dataset clearance to defend against such claims.
- “We assumed our old content rights covered AI training.” Pre-2023 creator contracts almost certainly did not contemplate AI training use cases. Brands that fed historical creator content into AI pipelines without renegotiation now face unquantified liability.
- “We let the agency handle disclosure.” As discussed earlier, brands bear primary FTC liability for disclosure failures, regardless of contractual delegation to agencies.
Sozee Workflow: Locked Likeness and Reusable Assets
Virtual-influencer licensing creates an operational challenge: teams must maintain character consistency across every asset while meeting disclosure and consent requirements at publication. Sozee addresses this at the platform level rather than relying on contract language alone.
The Sozee workflow maps directly onto the licensing controls described in this guide:
- Cast. Upload three photos to reconstruct a real likeness, or use the AI Character Builder to generate an entirely original character, a face that has never existed, with no source-person exposure. Compliance and verification sit inside setup instead of arriving as an afterthought.
- Direct. Photo Control locks five dimensions, Setting, Outfit, Shot style, Expression, and Object, so the licensed character appears identically across every asset in the deliverable. Locked likeness serves as the platform’s core consistency guarantee and reduces “real-person resemblance” risk when AI outputs might otherwise drift.
- Create and Reuse. Every setting, outfit, and object becomes a saved, reusable asset. A brand that licenses a virtual influencer for a 12-month campaign builds the character’s world once and reuses it across every activation, with no re-prompting and no consistency drift.
- Publish. The Scheduler connects to Instagram, TikTok, X, Facebook, Reddit, and Fanvue per character. It applies platform-specific captions and disclosure language before publication, which supports the real-time monitoring requirement that contracts alone cannot satisfy.
- Measure. Analytics separate Sozee-posted content from manually posted content. Legal and compliance teams gain an auditable record of what was published, when, and on which platform, the level of documentation that supports quarterly compliance attestations.
Because Sozee’s likeness models are private, isolated, and never used to train external systems, the platform’s architecture aligns with the data-isolation and training-prohibition clauses recommended throughout this guide.
2026 Right-of-Publicity Outlook for AI Personas
The legislative trajectory for 2026 and beyond points toward federal preemption layered over a durable state-law floor. If the NO FAKES Act passes the full Senate and is signed into law, it will establish a nationwide licensable property right over voice and visual likeness while preserving existing state causes of action. Brands will then face both federal liability and state-law claims in jurisdictions such as California, Tennessee, and New York simultaneously.
The virtual influencer market reached $11.74 billion in 2026 and is projected to reach $154.6 billion by 2032 at a 41.29% CAGR, which magnifies the commercial stakes of unresolved licensing disputes each year. Brands and agencies that standardize their likeness license agreements now, with explicit AI training, derivative, and disclosure clauses, will be positioned to operate in whatever regulatory environment emerges from the current legislative cycle.
Build your federal-and-state-compliant campaign with Sozee’s locked-likeness platform.
Frequently Asked Questions
What is the difference between a usage right and a training right in a virtual-influencer license?
A usage right permits the licensee to reproduce and distribute the virtual influencer’s likeness in defined campaigns, on defined platforms, for a defined period. A training right is a separate grant that permits the licensee to input the character’s visual assets, voice samples, or behavioral data into an AI or machine-learning model to train, fine-tune, or validate that model.
The two rights carry different economic values, risk profiles, and statutory treatment under 2025–2026 state laws. Every virtual-influencer license agreement should address them in separate clauses with separate compensation structures. Silence on training rights is not a neutral position and instead creates unresolved liability.
Does a fully AI-generated character with no real-person source still require right-of-publicity compliance?
Yes, brands still face right-of-publicity and disclosure obligations. If the AI-generated character’s appearance is substantially similar to an identifiable living or deceased individual, even unintentionally, the brand faces right-of-publicity exposure under state statutes in California, New York, Tennessee, Montana, and others.
If the character appears in advertising distributed to New York consumers, New York S.8420-A also requires a conspicuous disclosure that the performer is AI-generated, regardless of whether any real person’s likeness was used as a source. Brands must document independent creation, maintain records of human creative decisions in character development, and apply platform-level AI disclosure labels before publication.
What disclosure language satisfies both FTC and New York S.8420-A requirements for a virtual influencer post?
The FTC requires disclosure of both the material connection between the brand and the virtual influencer and the fact that the endorser is not a real person. New York S.8420-A requires a clear and conspicuous statement in the advertisement itself that the performer is AI-generated.
Disclosure placed only in a bio, at the end of a caption, or through an ambiguous hashtag such as #partner does not satisfy either standard. A compliant disclosure appears at the beginning of the content, is visually prominent, and uses unambiguous language such as “AI-generated character” or “virtual influencer.” Brands, not agencies or creators, bear primary enforcement liability for missing or insufficient disclosures under the current FTC enforcement posture.
How long should a virtual-influencer likeness license last?
Current practitioner guidance favors time-limited grants with renewal options instead of perpetual grants. Perpetual AI likeness rights are increasingly difficult to enforce and carry reputational risk because the economic value of a digital twin grows with technological improvements over time.
A license signed in 2023 under broad “future technology” language may now cover AI training and derivative uses the licensor never contemplated. Time-limited grants allow both parties to renegotiate compensation based on demonstrated usage and market rates at renewal, which aligns incentives and reduces the likelihood of mid-term disputes.
How does Sozee’s locked-likeness architecture reduce licensing disputes?
Most virtual-influencer licensing disputes arise from character inconsistency across campaign assets, unauthorized derivative outputs, and undocumented usage. Sozee addresses all three at the platform level.
Locked likeness ensures the licensed character appears identically across every image, video, and publication in a campaign, which removes the drift that creates “real-person resemblance” exposure when AI outputs vary. Every setting, outfit, and object is saved as a reusable asset tied to the specific character, creating an auditable record of what was produced.
The Scheduler applies platform-specific disclosure language before publication and generates a timestamped record of every post, which supports the quarterly compliance attestations that well-drafted licenses now require. Because Sozee’s character models are private, isolated, and never used to train external systems, the platform’s architecture remains structurally compatible with the training-prohibition and data-isolation clauses recommended in this guide.