AI avatar consent release checklist for creators and teams

Start with a specific permission question

A photo you can upload is not necessarily a photo you can turn into a spokesperson. A recording you can edit is not automatically permission to clone its speaker’s voice. Before collecting signatures, describe the transformation in terms the person can understand: which portrait will move, whose speech will drive it, what words it will appear to say, and where viewers will see the result.

This guide helps a producer assemble that record and identify questions for a qualified reviewer. Rights, enforceability, labor agreements and disclosure obligations depend on the project and jurisdiction. Do not present the record below as a legally sufficient release everywhere. For a jurisdiction-specific overview of voice rights, start with the linked voice-cloning guide; this page focuses on the actual approval handoff.

Use three separate approvals: permission to use the inputs, permission to create the described synthetic portrayal, and permission to publish the selected output. Combining these into one vague “AI use” checkbox makes it difficult to distinguish a private test from a paid campaign.

Related rights guidance: AI voice cloning: consent, licensing and disclosure.

1. Identify the person, voice and source assets

Record the subject’s name or your internal participant identifier, the project owner, and the person authorized to grant each permission. Keep sensitive contact and identity documents in your controlled production record, not in public captions or shared example folders. This checklist assumes an adult participant; projects involving children require a different review rather than casually adapting an adult form.

For each portrait, record its filename, creator or licensor, source, license reference and permitted edits. For each recording, identify the speaker, recording owner and whether it is uploaded as the final performance or used to create a reusable synthetic voice. A photographer’s permission and the pictured person’s permission answer different questions; do not use one as evidence for the other.

Write down any third-party music, stock footage, visible trademarks or artwork used in the finished clip. Keep their permissions separate from the avatar release so replacing a soundtrack does not silently change the scope approved by the participant.

2. Describe the synthetic operation plainly

Say whether the project animates a still photo, changes a recorded person’s mouth movement, swaps a face, generates new speech or creates a reusable voice. State which operations are authorized and which are excluded. “Make a talking portrait from this photo and this recorded script” is more informative than “use AI on my content.”

Include a short representative preview if the person needs it to understand the intended portrayal. Label that preview as an example for approval. Permission to inspect a private test should not be treated as publication permission. Document who may see previews, where they are stored, and how the final version will be approved.

Explain the boundaries of editing: cropping, captions, translation, voice replacement, changes to emphasis, facial motion or insertion into a new scene. If you plan to let another agency generate future versions, name that party and define its scope. Do not presume an approval transfers to unlimited future operators.

3. Scope scripts, channels, languages and duration

Attach the approved script and version number. Specify whether the person must approve every future script, may approve a bounded series of educational scripts, or has granted some other reviewed scope. Keep factual and endorsement claims attached to the words actually approved, not merely to the visual likeness.

List owned websites, help centers, social accounts, paid advertising channels and internal training separately. Record audiences, territories, languages, campaign dates and any permitted reuse. A consent record can specify a new approval for translation because a participant may approve the English script without being able to verify a localized statement.

Define compensation and responsible payment party, including whether the permission covers one deliverable or a continuing series. Employment, participation in a photo shoot or a tool subscription should not be used as your production team’s substitute for a documented synthetic-use decision. Ask your reviewer what the underlying agreement actually permits.

4. State prohibited portrayals and required approvals

Invite the participant to identify contexts they do not authorize. Examples to discuss include political messaging, medical endorsements, financial promises, intimate portrayals or statements framed as personal experience. These examples are a scoping aid, not an exhaustive list of prohibited uses or a claim about every country’s law.

Name the person who approves the final portrayal and the channel used for that approval. Define whether a missing response means the project remains on hold; do not let a deadline silently turn silence into permission. Retain the exact output reviewed, including audio, subtitles, thumbnail and surrounding advertising copy.

Set reapproval triggers. A different product claim, new language, paid placement, changed voice, new audience or materially different context can exceed the original production scope. Keep a small change log tied to the agreement so a teammate can tell a formatting correction from a substantive new use.

5. Plan revocation, retirement and access

Record how a participant can raise a concern, who receives it and who can pause distribution. The agreement and applicable law determine what withdrawal means; this checklist does not establish a universal right to revoke or a universal rule that permission cannot be revoked. Ask the reviewer to specify treatment of future generations, current campaigns and already distributed copies.

Separate actions your team controls from those it cannot guarantee. You may be able to stop new generation, remove an owned page or ask an agency to retire an ad. You cannot promise that every downloaded or third-party copy disappears unless you can actually deliver that promise. Record takedown requests and completion evidence individually.

Assign access to the people who need the inputs and approvals. Choose a retention and deletion procedure that matches your contracts and actual tools. Do not promise a vendor-specific deletion deadline from memory or describe all processing as local when files are sent to a hosted service.

Copyable production approval record

Use this as a briefing template for your contract or approval process. Fill every field with a concrete value or an explicit “not authorized”; do not leave future reuse open by omission.

Project and version: [campaign name / record ID / revision date]. Producer and contact: [responsible party]. Participant and authorized representative: [identity / capacity]. Source assets: [photo and recording filenames / owners / permission references].

Synthetic operations allowed: [talking-photo animation / lip synchronization / voice generation / other]. Operations excluded: [specific exclusions]. Script and language versions: [attached document IDs]. Review preview: [file ID / restricted access / review-only status].

Publication scope: [named channels / internal or public / paid or unpaid / audience / territories / languages]. Term and reuse: [start / end / permitted archive or reuse]. Compensation: [amount or arrangement / payer / covered deliverables].

Approval procedure: [reviewer / required confirmation / exact export ID]. Reapproval triggers: [new script, translation, channel, voice or context]. Prohibited contexts: [agreed restrictions]. Disclosure plan: [on-screen wording / caption / platform control / reviewer].

Access and retention: [operators / vendors / storage / retention / deletion procedure]. Concern or withdrawal process: [contact / response owner / agreed consequences / treatment of existing copies]. Permission evidence: [signed agreement or reviewed approval record / date / version].

Keep the record beside the approved source and final export. Preserve an immutable copy of the approved version; a link to a document that can change later is insufficient to show what the participant actually reviewed. Limit record access without removing the information the authorized reviewer needs.

Worked example: a founder’s product-education avatar

This is an illustrative planning example, not a customer story or a tested legal release. Suppose a founder wants a 20-second onboarding clip explaining how to upload a photo. The input is a licensed portrait and the founder’s own recorded narration. The producer proposes talking-photo animation, captions and a square crop.

The proposed scope is the company’s help center and owned onboarding messages in English. Paid ads, voice cloning, translated speech and future testimonials are excluded until separately approved. The founder reviews the complete clip and the help-center text around it, then approves that exact export. The team stores the recording, script, permission evidence and export under one record ID.

Later, the marketing team wants to reuse the avatar in a paid advertisement stating “customers save hours every week.” That is a new claim and distribution context. The existing education approval is not treated as approval for the ad. The team obtains a reviewed new scope and substantiates the claim, or changes the creative. It does not solve the gap by adding a generic AI disclosure.

For an employee training avatar, adapt the audience and access scope to the actual training system. For a performer in localized ads, attach every language version and the relevant campaign scope. These are separate approval designs, not evidence that one signed template covers all three.

Disclosure and copyright answer different questions

For U.S. marketing, the FTC’s Endorsement Guides Q&A addresses truthful endorsements and disclosure of material connections. Check its current guidance when an avatar presents a paid recommendation or testimonial. A statement that content is AI-generated does not by itself explain an undisclosed advertiser relationship or make a fabricated endorsement truthful.

The U.S. Copyright Office treats digital replicas and the copyrightability of AI outputs as distinct topics in its AI reports. Keep input permissions, portrayal permissions and any copyright question separate in your record. Do not label an output “copyright protected” simply because your team paid for generation or obtained the subject’s approval.

Review the destination’s current synthetic-media controls and the actual risk of viewers confusing the clip with a recorded performance. Agree on plain wording such as “AI-animated portrait using an approved recorded script” when that description is accurate. Have a reviewer check the label, placement and any additional legal or platform requirements for the specific publication.

How this applies to Magic Hour

Magic Hour’s Talking Photo tool accepts a portrait and audio to create a speaking video. Obtain permission for both inputs and the proposed portrayal before uploading; inspect the whole generated clip before seeking final publication approval. The tool does not supply a participant release or verify your rights for you.

Under the current Terms, commercial use requires a paid subscription, a credit-pack purchase or a separately agreed paid usage-based billing arrangement, and remains subject to third-party rights. The Terms and Privacy Policy state that customer prompts, uploads and generated assets are not used to train AI models. Authorized processing or access to provide the service and address specified operational needs is separate from a claim that no vendor or person can ever access a file.

Keep a dated reference to the policies you reviewed. Do not copy old competitor policy language into your participant record or promise security, retention or legal guarantees the service does not make. If your project’s requirements are stricter than the documented service behavior, resolve that gap before generating.

Final approval before upload and publication

Before upload: identify the participant and input owners; record the exact synthetic operation; check source permissions; agree on access and storage. A test permission should be labelled as a test, with its permitted viewers and use.

Before publication: match the final script, voice, likeness, context and channels to the approved scope; obtain confirmation for the exact export; verify factual claims and endorsements; check disclosure and destination requirements. If a field remains unknown, keep that particular use on hold rather than guessing.

After publication: retain the approved version and where it was distributed, route concerns to the named owner, and reassess the record when someone proposes a new message or campaign. This makes future reuse faster because the team can see what is already permitted and what needs a fresh decision.

Official sources and related workflows

FTC endorsements, influencers and reviews

U.S. Copyright Office: copyright and artificial intelligence

Magic Hour Terms of Service

Magic Hour Privacy Policy

Talking Photo input and workflow reference

Compare talking-photo workflows

Advertising consent and disclosure checklist