When AI recognizes your digital double

6 min read

A face can become a replica able to act, speak and circulate without the person it came from. The problem is not only recognizing the fake—it is deciding who controls identity.

Photographs, footage or a short scan can be enough to turn a face into reusable material. AI can reconstruct it, age it, make it younger, have it speak new lines or appear in scenes the person never shot. The face thus stops being only a feature of identity and becomes a productive asset: a digital double that can work even when its original is absent. The most urgent question is no longer whether the replica looks real, but who has the right to create it, store it and make it act.

The U.S. Copyright Office defines a digital replica as an image, voice or other digitally generated or manipulated representation capable of realistically depicting a person. Not every synthetic face is therefore a stolen face: an invented character, an authorized stand-in or an agreed transformation can serve a legitimate creative function. The problem begins when an identifiable person is reproduced without specific consent, or when permission granted for a limited use is extended to contexts, territories or time periods that were never accepted.

In cinema, this distinction enters directly into the workflow. An actor may authorize a scan to complete a difficult scene, but that does not mean permanently handing their identity over to a studio. Archival photographs, screen tests, promotional material and images published online can also become sources for training or guiding a replica. The boundary between document and performer becomes thinner: the same image that yesterday served as reference material can today become the basis of a presence capable of acting. Consent to collect data should therefore never automatically become consent to any future performance.

The contractual protections developed by SAG-AFTRA insist on precisely this point: creation and use of a replica are separate acts, and the use must be clearly described. Specific consent concerns the project, function, duration and compensation; a new use may require a new authorization. This is important for extras and small roles as well, whose faces could be multiplied to fill a synthetic crowd. Technology makes duplication cheap, but it should not make labor invisible or turn one day on set into a permanent license.

Transparency solves only part of the problem. Provisions of the European AI Act applicable from August 2026 require certain AI-generated or manipulated content to be identifiable and deepfakes to be disclosed. Knowing that an image is synthetic helps the public interpret it, but a label is not an authorization. The words ‘generated with AI’ do not establish who approved the use of a face, how much they were paid, whether consent can be revoked or whether the replica must be deleted at the end of the project. Transparency for the viewer and control for the person are two different protections.

Copyright alone does not provide a complete shield either. A person’s face does not function like a protected work in the same way as a film or photograph. Depending on the country, image rights, privacy, publicity rights, unfair competition, contracts and anti-deception laws may all apply. The U.S. Copyright Office report describes this very mosaic of protections and proposes a specific federal safeguard against unauthorized digital replicas. As long as the rules remain fragmented, an international production must check not only where the content is created, but also where it is distributed and which rights each territory recognizes.

Production therefore needs an identity register similar to those already used for music, stock assets and releases. It should connect every scan or source image to the person, the authorization, permitted uses, duration, territories and any limitations on the model. Access, new generations, delivered versions and deletion procedures should also be recorded. This traceability is not bureaucracy added after the film; it is part of production continuity, because it prevents a replica valid for one scene from being reused without control in an advertisement, sequel or training archive.

A digital double can be a powerful creative tool: it can make risky scenes safer, preserve continuity, facilitate localization, build impossible characters or complete a work with the agreement of the rights holder. But technical quality is not enough to make it legitimate. A synthetic face can be perfectly believable and still legally unauthorized. The question to ask when confronted with that presence is not only ‘is it real?’ but ‘who authorized this face to be here?’

  • Digital double
  • Replica digitale
  • Diritti d’immagine
  • Consent
  • Deepfake
  • Identità sintetica
  • AI Act
  1. Digital Replicas Report — U.S. Copyright Office
  2. Copyright and Artificial Intelligence — U.S. Copyright Office
  3. Risorse sulle repliche digitali — SAG-AFTRA
  4. Consenso per le repliche digitali — SAG-AFTRA
  5. Diritti sull’immagine digitale — SAG-AFTRA
  6. AI Act e trasparenza — Commissione europea
  7. Deepfake e identità — EU Intellectual Property Helpdesk