FilmFestival Cottbus begins on 3 November 2026; two days later, Sandra Prechtel’s documentary “Einar Schleef – Ich habe kein Deutschland gefunden” (Einar Schleef – I Have Found No Germany) opens in German cinemas. Their proximity raises a question that goes beyond individual films: who bears responsibility when a voice in cinema is no longer simply recorded, but generated or altered using artificial intelligence?
The distinction is crucial. According to information about the film, Prechtel’s portrait allows Schleef to speak for himself through archive material. A historical recording of this kind is not the same as a synthetic voice that produces new sentences in the voice of someone who has died. The latter requires a different kind of justification – and contextualisation that is clear to the audience.
A sound is not consent
AI can use voice recordings to build a model that generates new utterances. In film, applications range from technical corrections to replicating a specific person. The result may sound plausible; but a plausible sound proves neither that the person spoke those words nor that they consented to such use.
The technology therefore raises more than a question of labelling. A voice is part of a person’s self-presentation as well as the artistic work of actors and voice artists. Consent should therefore not be understood as blanket permission for any subsequent use. What matters is the purpose, scope and duration of the use – and whether a new use requires fresh consent and remuneration.
For actors in German film and television productions, collective agreements contain rules on generative AI. These concern the digital replication and use of their likeness. This is a concrete step, but not a universal answer: dubbing artists, documentary teams and people whose voices survive only in old archive material each operate under different conditions.
Labelling does not confer legitimacy
Transparency obligations under the European AI Act have applied since 2 August 2026. For certain AI-generated or manipulated content that could be mistaken for a genuine recording, the artificial alteration must be disclosed. For artistic and fictional works, the form of disclosure must be chosen so that it does not unnecessarily compromise the work. However, this replaces neither the consent of the person concerned nor the need to clarify rights to the recordings used.
Labelling and consent thus address different problems. A notice can tell audiences that a voice is synthetic; it does not automatically explain who authorised its use. Conversely, contractual consent may be in place without viewers realising how extensively a voice has been altered. Particularly in documentary film, which depends on its relationship to reality, it should therefore be clear whether audio comes from an archive, has been processed or has been newly generated.
The archive speaks – but not without limits
According to filmportal.de, Prechtel’s film about Einar Schleef, who died in 2001, assembles archive material and lets the artist speak for himself. This does not mean that every archived utterance is freely available for use. Using recordings can also implicate rights; a newly generated statement would, moreover, be an attribution that audiences could easily take to be authentic.
The dividing line is not simply between analogue and digital. A historical recording, too, involves selection and editing; a synthetic voice, too, can be a legitimate creative device in an openly fictional work. What matters for credibility is whether a film makes it possible to understand what has been documented and what has been constructed. And for responsibility, whether the people whose voices become material are appropriately involved.
In Cottbus, festival audiences and a cinema release now intersect with a debate that no single technical decision can settle. The central question is not merely whether a voice sounds artificial, but who determines how it is used – and whether audiences are told what they are hearing.
Original reviewed by the editors · 3 November 2026
