Get Consent Before Synthetic Voice or Likeness in a Project
Beginner5 min readCreative Craft with AI

Get Consent Before Synthetic Voice or Likeness in a Project

Project hygiene for makers who want synthetic voice, face, or performance elements: consent first, scope in writing, storage limits, and clear refusal paths - without a cloning how-to.

What you should be able to do

Synthetic voice or face in a project is a consent and scope problem before it is a technical one. Get a clear yes for a defined use, or do not generate.

AI Expert TeamPublished: Jul 31, 2026
Saved only in this browser.
In this article

It is getting easy to put a familiar voice on a new script or a familiar face on a new scene. Ease is not permission. This article is project hygiene for makers: how to decide whether synthetic voice or likeness belongs in the work at all, and what consent has to cover before you generate.

It is not a cloning tutorial. For personal scam defense around your own voice, use voice clone personal safety check. For still photos of people, use photo consent before AI edit or share. For adult non-consensual intimate imagery emergencies, use adult deepfake first response and StopNCII.org. Stage-level taste decisions remain in keep your taste.

Why “they would probably be fine with it” fails

Illustrative scenario: a podcast producer clones a co-host’s laugh as a transition sting “as a joke” using old episode audio. The co-host hears it in the published feed. Even if the technical quality is imperfect, the trust break is perfect.

Consent is specific. A person who agreed to be recorded for Episode 12 did not automatically agree to a synthetic performance in a trailer, an ad, or a fictional scene.

Do not build a workflow that starts with generation and asks for forgiveness. Synthetic media is hard to claw back once distributed. Consent is a gate before render, not a caption after upload.

Before any synthetic voice or likeness generation for a project, get a clear yes that covers:

  1. Who - the identifiable person (or rights holder for estates / minors - get appropriate guardian/authority involvement; do not improvise).
  2. What - voice only, face only, full performance, stills, etc.
  3. Which scripts/scenes - attach or summarize the intended lines and contexts.
  4. Where it will appear - private draft, password review, public feed, paid ads, training data, merchandise.
  5. How long - one campaign, one season, perpetual? Prefer time-bounded rights.
  6. Revocation - how they can say stop, and what you will take down.
  7. Compensation - unpaid cameo vs. paid performance; write it down.
  8. Storage - where reference audio/video will live, who can access it, when it will be deleted.

Put the answers in email or a short release - not only in a chat thread with a model.

Reference recordings are sensitive biometric-adjacent material in practice even when your local law uses different vocabulary. Store them in access-controlled folders, do not paste them into random consumer accounts, and delete them when the license ends. Pasting a client’s voice sample into an unapproved tool can also be a work-secrets problem (do not paste work secrets).

Project hygiene steps

  1. Decide if you need likeness at all. A narrator hire or stock licensed for commercial use may be cleaner.
  2. If you proceed, freeze consent before procurement of tools. Tool choice is secondary.
  3. Generate only inside the approved scope. New scene = new yes.
  4. Label synthetic performances in credits when audiences could assume a real performance (disclose AI in creative credits).
  5. Keep a kill switch. Know how you will remove the asset from feeds and ads if consent is revoked.

Special-care cases

  • Minors: default to no synthetic likeness for public projects without specialized legal and parental/guardian process. This article will not soften that.
  • Intimate or sexualized content: require explicit, enthusiastic, documented consent for that exact use - or do not make it. Non-consensual intimate imagery is a safety emergency, not a creative experiment (StopNCII; adult first response).
  • Deceased persons: rights and cultural norms vary; do not treat “they’re gone” as a free license. Get counsel and family/estate clarity when the use is public or commercial.
  • Employees and freelancers: company recording policies rarely equal marketing-clone rights. Get a project-specific yes.

EU transparency rules for certain deepfake-like content intensify under AI Act Article 50 from 2 August 2026 (Regulation (EU) 2024/1689; see Commission FAQ). If you put a synthetic voice or likeness into your own project you are normally a deployer, so the Article 50(4) labelling duty applies to you in full from that date; the limited marking grace to 2 December 2026 covers only providers of generative systems already on the market, and is no help to you (Regulation (EU) 2026/1744, inserting Article 111(4)). Consumer-protection agencies have also warned about voice clones used for deception (FTC on preventing harms from AI-enabled voice cloning). Separately, copyright offices still center human authorship of claimed expressive elements (Copyrightability Report, Jan 2025), and WIPO’s generative-AI IP materials stress clearance and consent rather than casual confidence (WIPO, Generative AI: Navigating Intellectual Property). Your project can be non-fraudulent and still be unethical if consent was skipped - aim for both lawful and decent. This section is project hygiene, not legal advice.

I agree that [Producer] may create a synthetic voice performance using my
provided reference audio for [Project], limited to the script dated [date],
for publication on [channels], until [end date]. I may revoke future use by
email; [Producer] will remove the asset from controlled channels within
[N] days. Reference audio will be stored in [location] and deleted by [date].
Compensation: [terms]. This is not consent for intimate, political, or
defamatory uses, or for training a public model on my voice.

This is an illustrative project-hygiene template, not a lawyer-vetted release for every jurisdiction. High-stakes commercial uses deserve counsel.

Refusal and redesign patterns

When consent is missing or partial:

  • Recast with a consenting actor
  • Use original narration you perform yourself
  • Redesign the scene so identity is not required (silhouette, abstract audio)
  • Drop the bit

Do not “anonymize” a voice that is still recognizable to anyone who knows the person. Recognizability is the test that matters ethically.

Before your next project that “could use a quick clone,” run the voice and likeness consent checklist. If any row is blank, the technical demo does not start. Hire a consenting performer or redesign the scene.

Read next

Continue through the same learning path with the next practical articles.