Skip to content
Larping Agency Subscribe

Check the Rights Before AI Turns Your Content Into a New Performance

Devon Ariza

A UGC contract may cover some AI-assisted edits, but neither the agreement’s title nor a generic commercial-use grant settles the question. Coverage depends on the exact rights granted, the proposed alteration, where and how the result will be used, the contract term, and the governing law.

Silence about AI creates ambiguity—not automatic permission and not an automatic ban. If broad legacy wording is the only apparent authority for changing the creator’s identity, conduct, endorsement, or performance, pause and confirm the proposed use in writing before publication.

The short answer: use a green, yellow, or red test

Start with the proposed output, not the word “AI.” Background cleanup and a fabricated testimonial may use similar technology, but they ask the contract to authorize very different results.

Result What the contract and use look like Next step
Green Express, specific permission covers this alteration, asset, campaign, medium, territory, term, vendors, and downstream use. Required approval has been obtained. Record the approval and publish only within scope.
Yellow The agreement allows content to be “altered,” “transformed,” “composited,” used in “derivative works,” or modified broadly, but does not define AI, synthetic media, training, replicas, or generated performances. Obtain written clarification or an amendment.
Red The contract reserves or prohibits AI use; the output enters a forbidden context; the product, campaign, platform, territory, account, or term is outside scope; or use continues after expiry. Stop and escalate before generating or publishing.

A yellow result is not a loophole for either side. The creator should not assume silence blocks every ordinary edit, and the brand should not assume “modify” authorizes a new pose, claim, endorsement, or performance. The practical fix is a written description of the alteration, supported by the proposed or representative output and approved by parties with authority to change the deal.

Escalate immediately if AI would realistically change the creator’s identity or apparent conduct, produce a synthetic testimonial, place them in sexualized, defamatory, political, discriminatory, or deceptive content, or connect them to regulated-product claims. Perpetual rights, broad sublicensing, reusable replicas, biometric data, and post-expiration use also warrant prompt review. Lathrop GPM’s guidance recommends separating routine editing from synthetic recreation and defining purpose, media, territory, duration, vendor access, approval, and prohibited contexts rather than relying on broad legacy wording (Lathrop GPM).

Separate routine editing from a performance the creator never made

Use a three-level ladder to identify what the contract is actually being asked to permit.

Level one: routine post-production. Cropping, resizing, color correction, noise reduction, and background cleanup resemble conventional editing more closely than synthetic performance. They may fit an ordinary post-production clause, although the contract still controls. Permission for color correction does not quietly become permission to redesign a face or body.

Level two: material alteration of an authentic image. This includes changing the creator’s face, body, clothing, expression, setting, or apparent conduct. Replacing a messy room with a neutral studio is not equivalent to changing the creator’s expression to imply excitement, altering their clothing, or placing them at an event they never attended. The original image may remain recognizable while the result communicates something the creator did not do.

Level three: a newly generated performance or replica. Examples include a new pose, spoken statement, testimonial, endorsement, scene, or reusable avatar that the creator never performed. This asks whether the agreement permits creation and commercial use of new conduct—not merely whether pixels in a delivered image may be retouched.

The U.S. Copyright Office describes a digital replica as a digitally created or manipulated image, video, or audio recording that realistically but falsely depicts an individual. In 2024, the Office recommended federal protection against knowing distribution of unauthorized digital replicas. That was a policy recommendation, not enacted law (U.S. Copyright Office).

Keep the human subject separate from the environment. AI used only for lighting, sets, backgrounds, cleanup, or similar environmental elements may fit narrow production permission more readily. AI applied to a creator’s face, voice, body, identity, or performance requires a separate and more exact review.

Audit the contract clause by clause

Search the agreement and every incorporated exhibit, brief, release, amendment, purchase order, and platform authorization for:

AI, generative AI, synthetic media, digital replica, manipulated likeness, avatar, model training, modify, alter, adapt, composite, transform, derivative works, raw materials, sublicense, assign, paid media, and whitelisting.

Do not stop when one promising word appears. Read definitions, exclusions, survival provisions, and any rule governing conflicts between the main agreement and campaign brief.

Clause Question to answer Red flag
Ownership and identity Who owns or licenses the deliverable, and what rights cover the creator’s name, image, voice, likeness, handle, and persona? “Work for hire” or assignment is treated as automatic permission to synthesize the person.
Editing and creation May the brand edit delivered assets only, or generate new images, poses, speech, and performances? Unlimited derivatives with no distinction between retouching and replicas.
Scope and duration Which product, campaign, media, platform, account, territory, and term are covered? Perpetual worldwide use across unidentified products or future campaigns.
Data and vendors May raw files, scans, recordings, or biometric information go to AI tools, agencies, affiliates, or vendors? Training, retention, sublicensing, and deletion are unaddressed.
Control and exit Are approval, correction, takedown, revocation, fees, termination, and deletion defined? No remedy for misleading output or continued use after expiry.

Review ownership or licence language separately from identity rights. A brand may own a photograph or receive an assignment of copyright without necessarily receiving permission to build a synthetic likeness. Likewise, a licence to publish delivered assets may not extend to generating images that never existed. Contract guidance on digital replicas therefore recommends treating ownership, likeness alteration, generated performance, and training as distinct permissions (Lathrop GPM).

Ask whether “derivative works” applies only to edits of supplied files or also to scenes generated from facial references, raw video, scans, or voice recordings. Check whether access to “raw materials” includes the right to upload those materials to third-party systems.

Then map the commercial scope: named product, campaign, organic or paid media, platform, account, territory, term, exclusivity, archival display, and renewal. Whitelisting or allowlisting—advertising through a creator’s account—is an account-level permission distinct from permission to use an image.

Treat the following as separate rights rather than one general “AI permission”:

  • Accessing raw files, facial scans, voice recordings, or biometric information
  • Uploading materials to an AI provider
  • Training or fine-tuning a model
  • Creating embeddings or a reusable avatar
  • Generating a new asset or performance
  • Distributing the output
  • Sublicensing material to agencies, affiliates, platforms, or vendors

Finally, inspect approval, correction, takedown, revocation, compensation, termination, and deletion terms. Approval of the original cut may not cover later AI variants. A deletion duty covering campaign files may say nothing about retained inputs, embeddings, models, or vendor copies.

Why broad alteration language still leaves an AI dispute

Words such as “altered,” “transformed,” “composite,” and “derivative works” may support a brand’s authorization argument. The available evidence does not establish a universal rule that those words authorize every generative-AI use.

The Rainbow Shops dispute provides a useful example. Reporting quotes a release allowing images to be used “whether intact or in part, composite or distorted in character or form, cropped or altered, without restriction as to changes or transformations.” Rainbow argued that this language covered “photoshop, AI or any other technological enhancements.” That was the company’s legal position, not a final statement of law (The Fashion Law).

The creator alleged that photographs taken against a plain background were used to generate substantially different images, including a sexually suggestive depiction, and that use continued after the original agreement expired. Those are allegations, not findings of liability (PetaPixel).

On July 16, 2026, a New York judge denied the creator’s request for a preliminary injunction, allowing the disputed images to remain posted while litigation continued. According to reporting published July 29, the procedural ruling did not finally decide whether release terms associated with conventional alteration covered AI-generated poses or settings the creator never performed (AI Fashion Law).

The practical lesson is to test four questions separately:

  1. Modification: May the original asset be edited?
  2. New performance: May the creator be depicted doing or endorsing something they never performed?
  3. Context: Is the resulting setting or message permitted, particularly if it is sensitive or misleading?
  4. Duration: Was the output created, published, renewed, and retained while the relevant rights remained active?

A broad editing clause may strengthen an argument on the first question while leaving the other three disputed.

Run a scope and timeline test before deciding

Build a one-line timeline for the asset:

Original shoot → delivery → AI upload or processing → creator approval → publication → contract expiry → renewal or no renewal → archival display or takedown → deletion

Attach a date and responsible party to every stage. This catches a common mistake: concentrating on whether an edit was authorized while overlooking that it was generated, posted, renewed, or kept live after the permitted term.

Next, ask whether the agreement covers only the original campaign or also:

  • New products and product categories
  • New advertisements or future campaigns
  • Localization, translation, or dubbing
  • Merchandising and packaging
  • Paid media and landing pages
  • Other platforms, brand accounts, or creator accounts
  • Archival display after active promotion ends

An authorized type of alteration can still fall outside the deal if it is used in the wrong medium, territory, account, campaign, or period.

Do not collapse the workflow into one right. Ingesting creator data, training a system, creating a replica, editing an asset, distributing the output, and sublicensing it are separate stages. Permission at one stage does not necessarily provide permission at the next.

The AI provider’s commercial-use terms add another layer; they do not replace the creator agreement. They do not by themselves prove copyright ownership, exclusivity, likeness consent, or clearance of every source photo, logo, voice, location, or other input. A practical review should separate the tool licence, copyright and control, input and likeness clearance, campaign licence, and platform or disclosure rules (Synthetic Actor).

Editing permission may also coexist with questions involving publicity, privacy, biometric data, false endorsement, deceptive advertising, defamation, copyright, or synthetic-media labelling. Which rules apply depends on the governing law, the parties, the information collected, the workflow, and the publication context. Do not assume that a particular state law, union agreement, or platform rule governs every UGC deal.

What creators and brands should do when the language is unclear

If you are the creator:

  1. Save the signed agreement, exhibits, briefs, amendments, emails, messages, and approvals.
  2. Request the proposed output—or a close visual example—not merely “permission to use AI.”
  3. Identify the AI tool, agency, subcontractors, and vendors that will receive your material.
  4. Ask where the asset will run, from which account, for which product, in which territories, and for how long.
  5. Request written clarification before approving a material change to your face, body, voice, apparent conduct, endorsement, or performance.
  6. Confirm what happens to inputs, outputs, scans, embeddings, and reusable replicas when the campaign ends.

If you are the brand or agency:

  1. Describe the intended change specifically, including what will and will not be altered.
  2. Obtain informed written consent for that defined use rather than relying on “enhancement.”
  3. Limit vendor access and prohibit unapproved training, retention, and reuse.
  4. Record the approved asset, context, channels, accounts, campaign, territory, and term.
  5. Track expiration, renewal, takedown, and deletion duties.
  6. Reapprove materially different variants instead of treating one approval as permission for every future output.

Negotiation need not be all or nothing. The parties can permit routine background cleanup while requiring separate approval for facial, bodily, voice, expression, conduct, or performance changes. They can also prohibit training, fine-tuning, embeddings, and reusable replicas; set separate fees for different rights; bar specified sensitive contexts; and require deletion of source data and retained outputs where feasible.

Keep an asset-level rights record covering the original asset, signed release, approved alteration, AI tool, vendors, source permissions, accounts, media, territory, required disclosures, approval date, and expiration date. That makes renewal, takedown, and audit work far easier than reconstructing permission from scattered messages.

Prompt legal review is appropriate when the use creates a new endorsement or performance, involves sensitive content or biometric data, depends on perpetual or broadly sublicensable rights, or continues after expiration. It is also sensible when the parties disagree about whether “alter,” “transform,” or “derivative works” reaches a realistic synthetic depiction.

Do not reduce the answer to whether the contract contains the word “alter.” Match the exact AI use against the agreement’s editing, likeness, training, distribution, duration, approval, and termination terms. Narrow, express permission is the clearest result; broad legacy wording is a reason to pause and confirm the intended use in writing, especially when AI creates conduct, endorsements, or performances the creator never made.

This checklist is general information, not an individualized contract interpretation. Review the complete agreement—including incorporated documents and governing-law terms—and seek qualified advice before approving or publishing a material AI alteration.