Stop Letting Perpetual Buyouts Price AI Replicas at Zero

A perpetual UGC buyout can allocate $0 separately to AI-replica rights even when it authorizes synthetic uses carrying a proposed $5,000–$25,000 in statutory damages per unauthorized violation. The verdict is to price AI training, replica creation, and synthetic deployment separately rather than letting “in perpetuity, all media now known or hereafter devised” absorb them without a line item. The NO FAKES Act remained pending in the latest supplied legal record, so that damages range is a proposal—not current federal law—but it exposes a contract risk that already matters under state, labor, privacy, advertising, and existing identity-rights rules. Husch Blackwell describes the proposed federal liability and the bill’s pending status.

The Case for Taking the Larger Buyout

The received wisdom is commercially understandable. A creator deciding between a limited whitelisting license and a larger perpetual buyout may prefer more guaranteed money now. The brand gets predictable costs, avoids repeated renewals, and can move approved footage between platforms and formats without reopening the deal.

That can be a rational bargain when the grant covers familiar uses: editing the delivered asset, running it in paid media, adapting its dimensions, and keeping an organic post live. A broad license is not automatically abusive or invalid. Some creators reasonably trade future control for a sufficiently large guaranteed payment.

The problem is not breadth by itself. It is an undivided fee that conceals a qualitatively different permission: creating new speech, images, or performances from the creator’s identity. A higher flat number does not show whether any portion was paid for voice-model training, avatar creation, generated endorsements, vendor reuse, or future synthetic campaigns.

No supplied source establishes a standard market price for those rights. There is therefore no defensible universal AI fee to paste into every creator contract. The supported conclusion is narrower: the right should be identified, authorized, and priced separately so both parties know what was purchased. August 2026 contract guidance specifically recommended pricing “AI and replica rights separately from standard fees” and defining consent for digital replicas. Lathrop GPM discusses that contract response and the proposal’s committee advancement.

One Sentence Can Hide Four Different Grants

Consider the familiar clause fragment:

In perpetuity, in all media now known or hereafter devised, Brand may edit, adapt, reuse, and create derivative works from the Content.

Each part does different work.

“In perpetuity” removes the ordinary renewal point at which a creator could reprice expanded usage. It may also leave the brand holding source footage long after the original campaign ends.

“All media now known or hereafter devised” is designed to cover new distribution formats. Moving an approved video from one technical format to its successor is not the same act as generating a new performance, but the clause does not draw that boundary.

“Edit” and “adapt” can reasonably include cropping, captions, color correction, noise reduction, compression, and platform-specific reframing. They can also become disputed words when a tool changes dialogue, emotional delivery, product claims, or apparent conduct.

“Derivative works” concerns adaptations of an asset. It does not necessarily settle the separate identity question: may the brand build a model of the creator’s voice or face and use it to make performances the creator never recorded?

That distinction is the economic gap. A brand may own the delivered video or hold broad copyright permissions while still needing separate authority to simulate the person depicted. Music licensing presents the same separation: control of a composition or sound recording does not necessarily authorize a replica of the performer’s voice or likeness. Paul Hastings explains why identity authority must be diligenced separately from traditional copyrights.

Enter your buyout, scope, term, and expected replica uses; the calculator shows which structure wins for those inputs.

The comparison needs one legal qualification. AI training alone is not necessarily a NO FAKES violation. The proposal’s described focus is unauthorized covered replicas and their public display, distribution, transmission, communication, or other public availability. Training permission still belongs in the contract because it determines whether source material can become a reusable model, but the calculator’s damages comparison concerns resulting unauthorized replica uses.

The Bill Is Pending, but the Drafting Gap Is Present

As of the latest supplied legal update, published August 20, 2026, the NO FAKES Act had not been enacted. August advisories reported that the Senate Judiciary Committee advanced the bipartisan proposal, while the supplied updates differ on the exact June date. Committee advancement is supported; an exact date should be confirmed against the official congressional record before publication or contract reliance.

The revised proposal was described as creating a federal right to authorize certain digital replicas of an individual’s voice or visual likeness. Its protection would apply to individuals generally, not only celebrities, union performers, or creators with large audiences. The proposal also contemplated civil claims over unauthorized public dissemination and a notice-and-takedown process for qualifying online services.

Those provisions may change before enactment. Definitions, exceptions, remedies, licensing rules, preemption, platform procedures, effective dates, and amendments all require confirmation from the controlling bill text. A contract should not say federal law currently requires a particular AI fee, approval process, deletion procedure, or royalty. The supplied evidence establishes none of those universal mandates.

Pending status also does not create a legal vacuum. California Labor Code section 927 has been described as making certain digital-replica provisions unenforceable when they lack a reasonably specific description of intended uses and the individual lacks qualifying counsel or union representation. That is a defined contract-enforceability rule, not a universal requirement governing every creator deal.

New York’s synthetic-performer disclosure rule, effective June 9, 2026, addresses covered advertising featuring a digitally created human performer who is not recognizable as an identifiable natural performer. Disclosure to an audience is different from authorization by an identifiable creator. A label does not cure missing consent, and consent does not eliminate applicable disclosure duties. This advertising-law analysis explains the reported New York definition and disclosure trigger.

Applicable state publicity, privacy, biometric, labor, advertising, trademark, copyright, contract, and unfair-competition rules depend on the creator, production, data, locations, and campaign. Nothing here replaces advice on the current bill or a particular agreement.

Split Editing, Training, Generation, and Publication

A workable contract gives each stage its own permission instead of relying on “AI use.”

This structure allows a creator to approve ordinary editing without approving a synthetic performance. It also permits narrower AI arrangements. A voice performer could allow a dedicated model for pickup lines from an approved script but prohibit general-purpose model improvement. A creator could permit internal testing while barring public outputs. A brand could license one translated campaign without buying unlimited future dialogue.

The identity element should be equally specific: recorded voice, generated voice model, face, full body, movement, gesture, expressions, mannerisms, or a named avatar version. “Name, image, and likeness” may not explain which model can be created, what source material feeds it, or what the replica may say.

Scope should identify the campaign, product category, audience, media, platforms, territory, term, paid or organic distribution, output count, and exclusivity. Creative authority should address new dialogue, translated speech, emotional delivery, scenes the creator never entered, factual claims, dynamic responses, and combinations with another person’s voice or body.

Permission should also follow the files. Raw footage, isolated audio, transcripts, scans, facial maps, embeddings, model files, prompts, rejected outputs, and production metadata may pass through an agency or AI vendor. The contract should identify approved recipients, prohibit unapproved vendor training, set retention rules, and require downstream restrictions at least as protective as the creator-facing agreement.

Price the Reusable Capability, Not Just the Original Video

The production fee pays for making the original asset. Ordinary usage pays for exploiting that asset within an agreed scope. Replica capture and model creation produce a reusable capability, while deployment fees price what the brand subsequently generates and publishes.

Those are distinct commercial events even when the parties ultimately choose one invoice. A transparent deal can show separate amounts for production, ordinary usage, replica creation, paid media, each campaign or output, renewal, localization, expanded territory, and exclusivity. The supplied record does not provide standard prices for any of them.

Separate pricing also prevents a misleading negotiation. If a contract lists $5,000 for production and perpetual all-media rights but assigns no amount to AI training or synthetic replicas, the replica allocation is not automatically $5,000. It is undisclosed. The calculator treats that silent bucket as $0 because that is the amount expressly itemized for it—not because the entire agreement necessarily has no legal value.

The proposed $5,000–$25,000 damages range is a liability benchmark, not a suggested creator rate. Compensation should reflect the authorized project, output volume, media, term, territory, exclusivity, model control, and reputational risk. A creator cannot assume that requesting $25,000 is required or that receiving it makes every use lawful. A brand cannot assume that a large production fee supplies specific consent for unlimited replicas.

Approval can be priced and operationalized at several stages: the model prototype, approved scripts and claims, language versions, final outputs, paid placement, and new campaigns. The contract should say who may approve, whether email counts, and what happens when a response is late. Credit and AI labels should be handled separately because neither substitutes for authorization.

Historical Bill Terms Must Stay Labeled by Version

Analysis of the 2025 proposal described lifetime digital-replica rights as nonassignable while allowing written exclusive or nonexclusive licenses. It reported a general maximum license term of 10 years. For a person under 18, it described a five-year limit, court approval, and expiration when the person reached 18, with separate treatment for rights governed by a qualifying collective bargaining agreement. O’Melveny’s analysis reports those 2025 assignment, term, minor, and collective-bargaining provisions.

Those are historical, version-specific descriptions. They should not be copied into a 2026 contract as confirmed requirements without checking the latest official text.

The evidence also does not establish a universal federal right to revoke every license, mandatory royalties, independent counsel in every agreement, approval of every output, universal audits, or mandatory deletion at expiration. State law, union agreements, or negotiation may produce some of those protections. They should be attributed to the correct authority.

An agreement should distinguish expiration, termination for breach, negotiated revocation, suspension during a dispute, and withdrawal of a separate consent where permitted. Calling every exit “revocation” obscures whether the right actually exists and what happens to published ads, physical inventory, archives, backups, and vendor-held models.

The Carve-Out Must Reach Agencies and AI Vendors

A creator-facing promise is ineffective if raw audio is uploaded under vendor terms allowing general product improvement. The operational chain may include the brand, influencer agency, production company, editor, localization studio, AI provider, distributor, retailer, platform, and media buyer.

Each permitted recipient needs the same project, training, generation, publication, security, retention, deletion, and transfer limits. One party should verify consent before source material enters a tool. Another should maintain the authorization record, approved scripts, output log, platforms, territories, campaign dates, and removals.

Warranties should follow control. A creator can warrant authority to sign and address conflicts in creator-supplied material. The creator should not guarantee the behavior of a vendor selected by the brand. The brand likewise needs vendor contracts that support any security, deletion, or non-training promises it makes.

At expiration, the agreement should separately address raw recordings, isolated stems, scans, biometric identifiers, embeddings, dedicated models, prompts, generated outputs, archives, and backups. Possible treatments include return, deletion, access termination, model retirement, restricted legal retention, and certification covering subprocessors. Absolute deletion should not be promised where a system cannot deliver it.

Published material needs its own runout rule. Organic posts might remain visible but lose paid promotion; active ads might stop immediately; a nonpublic legal archive might survive. Whatever the deal allows should have a defined duration and any additional fee.

Machine-readable provenance can support these records without replacing them. A Creator Assertions Working Group draft describes a C2PA consent assertion linking an asset to an external authoritative consent record while distinguishing identity consent from asset consent. The assertion is informational—not itself a license, warranty, ownership ruling, or enforcement mechanism. The CAWG draft explains that consent-reference model and its limits.

The Better Clause Preserves Ordinary Usage

A focused carve-out does not need to block routine post-production or make every crop a new negotiation. It can preserve the agreed organic and paid-media license while reserving model development and synthetic performance:

Brand may edit the delivered Content for formatting, length, captions, sound cleanup, color correction, and placement within the stated campaign. No license is granted to train or improve a model using Creator’s voice, likeness, movement, or other identifying attributes; create a voice clone, avatar, digital double, or synthetic performance; generate dialogue or conduct Creator did not record; or authorize a vendor to do so. Any such use requires a separate written scope, fee, term, approved vendor list, output authorization, and end-of-use procedure.

That is drafting language for discussion, not a universal template. Definitions and enforceability depend on the original agreement and applicable law. A narrow addendum can also be undermined by an older assignment, publicity release, affiliate license, or “future media” grant unless the amendment expressly controls conflicts.

The central negotiation is therefore not “AI or no AI.” It is whether the creator is licensing an existing asset, a new synthetic performance, or a reusable identity model—and whether the contract records a separate price for each. A perpetual buyout may still win when its scope is understood and its compensation is deliberate. It should not win by treating the highest-liability permission as an undefined extra included for $0.