If your product turns a short video of a person into an avatar that can say anything, you are handling something more sensitive than a profile photo. The person in the clip is often not your customer: it is a sales rep, a trainer, a creator or a founder whose face and voice your customer uploaded. Getting consent wrong is a trust problem first and a legal problem second, and both are much cheaper to solve in the product than after launch.
This article is not legal advice. It is a practical overview for founders and developers, written from a product perspective. Laws differ by country and US state, they are changing quickly, and how they apply depends on your facts. Talk to a qualified lawyer in the jurisdictions where you and your customers operate before you ship.
Why consent is a product problem, not just a terms problem
Most avatar platforms involve three parties, and only two of them signed your terms:
- You, the platform that stores the clip and renders videos.
- Your customer, who uploads the clip and writes the scripts.
- The person in the clip, whose likeness and voice are being reused.
A terms clause saying "you must have permission" is necessary, but it does not tell you whether permission exists. That evidence has to be captured in your product when the avatar is created.
The principles that hold up across jurisdictions
Design around principles that recur across privacy and publicity law, not around one statute. They are also what customers' legal teams ask about in procurement:
- Consent comes from the person depicted. Not from their employer, not from the account owner. If an employer uploads an employee's clip, the employee still has to agree.
- Consent is informed and specific. The person should know what the avatar will be used for (internal training, outbound sales, paid ads), for how long, and whether their voice will be cloned.
- Consent is recorded. Who agreed, when, to what scope, and which avatar it applies to.
- Consent can be withdrawn. There must be a working path to stop new videos and delete the avatar and the voice.
- Viewers are not misled. Where synthetic video could be mistaken for a real recording, label it.
What some current laws say
The rules below are examples, not a complete map. Check the linked sources for the current text.
EU: the AI Act's transparency rules
The EU AI Act (Regulation (EU) 2024/1689) includes transparency obligations in Article 50, which apply from 2 August 2026. Two parts matter for avatar products:
- Article 50(2) requires providers of AI systems that generate synthetic audio or video to ensure outputs are marked in a machine-readable format and detectable as artificially generated, where technically feasible.
- Article 50(4) requires deployers of AI systems that generate a "deep fake" to disclose that the content has been artificially generated or manipulated. The Act defines a deep fake as AI-generated or manipulated content that resembles existing persons and would falsely appear to be authentic.
Whether a given avatar video counts as a deep fake depends on context: a video of a real person speaking words they never recorded can easily qualify. European Commission guidance and law-firm summaries, such as this Baker McKenzie overview, discuss how the test is applied and note transition periods for some obligations. If your customers publish videos to EU audiences, assume disclosure will be expected and make it easy.
EU: data protection
Video and voice recordings of an identifiable person are personal data under the GDPR. Biometric data processed for the purpose of uniquely identifying a person is a special category under Article 9 with stricter conditions. Whether your pipeline falls into that category is a question for your lawyer, but either way you need a lawful basis, a retention policy and a deletion path.
US: state right of publicity and digital replica laws
There is no single federal right of publicity in the US. Rules come from the states, and several have added rules aimed at AI replicas:
- Tennessee's ELVIS Act (effective July 1, 2024) added voice, including a simulation of a voice, to the state's existing protection for name, image and likeness. It also creates liability for distributing a tool whose primary purpose is producing a specific person's likeness or voice without authorization. See Holland & Knight's summary.
- California Labor Code § 927 (from AB 2602) makes certain contract provisions allowing a digital replica of a performer unenforceable for new performances from January 1, 2025, if the provision lacks a reasonably specific description of intended uses and the person was not represented by a lawyer or union. See the code text.
- Federal: the NO FAKES Act would create a federal right over digital replicas of voice and likeness. It advanced out of the Senate Judiciary Committee in June 2026 but is not law as of that date. Check its current status.
| Principle | Where it shows up | What to build |
|---|---|---|
| Consent from the person | State publicity laws, GDPR lawful basis | Consent step tied to the person, not the account |
| Specific scope | California Labor Code § 927 | Use-case and duration fields on the consent record |
| Withdrawal | GDPR rights | Revoke flow that deletes avatar and voice |
| Viewer disclosure | EU AI Act Article 50(4) | Default "AI-generated" label in your player and exports |
Designing consent into your avatar flow
Capture consent from the right person
When an account creates an avatar of someone other than the account owner, send that person a link. They confirm their identity, read a plain-language summary of the uses, and agree. Only then does your backend call the avatar API.
A consent record worth keeping looks like this:
{
"avatar_id": "av_123",
"subject_name": "Dana Reyes",
"subject_email": "dana@example.com",
"scope": ["internal_training", "outbound_sales"],
"voice_clone": true,
"expires_at": "2027-10-15",
"agreed_at": "2026-10-15T09:12:00Z",
"consent_text_version": "v3"
}
Keep the exact consent text version so you can show later what the person actually agreed to.
Make withdrawal real
If the person withdraws or leaves the company, stop new generations immediately and delete the avatar. With PitchTwin, deleting the avatar also deletes the cloned voice:
curl -X DELETE https://pitchtwin.ai/api/v1/avatars/av_123 \
-H "Authorization: Bearer pt_live_your_key"
Then delete your own copies of the source clip and any stored videos that fall outside the remaining scope. Generated video and audio links from the API are signed and expire after 24 hours, so anything still circulating is in your storage or your customer's.
Label output by default
Because white-label output carries no watermark or vendor branding, the disclosure to viewers is your responsibility and your customer's. Ship a visible "AI-generated" label in your player and email templates by default, and let customers adjust wording rather than remove it. If you add machine-readable marking in your own pipeline, document how.
Say no to the hard cases
Some cases are not worth supporting in a B2B tool: minors, deceased people, public figures without a documented licence, and clips scraped from the web. Block them in your terms and in your upload flow.
How PitchTwin fits
PitchTwin builds each avatar from a real video of the person and keeps that footage, which is exactly why consent matters: the result looks 1:1 like them. Our Terms require every person in an avatar to have agreed to it, and the API gives you the deletion path. The consent capture, scope and labelling described above live in your product, where you know the person and the use case. More on the delivery model in white-label AI avatars and on how avatars are built in avatar from video.
Try it
Read the API docs to see how avatar creation and deletion work, then wire your consent step in front of POST /avatars. You can sign up and get 60 free seconds, no card required, to test the full flow with a clip of yourself.