Documentary cinema is built on an implicit promise: the images may be selected and edited, but they maintain a relationship with people and events that existed before the film. In Musk, presented by Alex Gibney at Venice, that promise is deliberately disturbed by an AI-generated avatar of Elon Musk assembled from his public statements.
Synthetic reconstruction belongs to a documentary tradition that already uses animation, actors, miniatures and manipulated archives to represent what was not filmed. The difference is perceptual. A photorealistic avatar carries the authority of the real face and may be mistaken for a recording. The more reconstruction resembles evidence, the more clearly the film must state its grammar. A discreet end credit is not enough if the sequence is experienced as documentation during the film.
The synthetic figure does not pretend to be unseen archival footage. It functions as a critical device, giving physical form to a public identity already constructed through interviews, posts, performances and contradictions. The deepfake is exposed rather than hidden; its artificiality becomes part of the argument.
Editing public statements is equally delicate. A subject may have spoken every sentence, but not in the sequence, tone or setting created by the film. An avatar adds expressions, pauses and looks absent from every original source. It creates a new performance attributed to a real identity. Accuracy therefore depends not only on authentic words, but on the viewer understanding that the body and situation are an authorial interpretation.
This does not remove the ethical problem. An avatar can combine words spoken in different contexts, alter rhythm and expression and create a performance the subject never gave. Even when every sentence is authentic, the body that delivers them creates a new meaning. Editing has always shaped reality, but synthesis adds a layer that resembles evidence while being newly manufactured.
Critical use should offer different levels of verification. A film can mark generated sequences, identify quotation sources and link digital audiences to original materials. Provenance credentials may document the file, but the editorial decision remains essential: explain why the avatar was necessary and which argument it constructs. Technology can certify a step; it cannot replace methodological honesty.
Legitimacy therefore depends on legibility. The audience should know that the figure is generated, which materials informed it and where reconstruction begins. Provenance, labelling and contextual explanation are not technical footnotes; they are part of documentary language. Without them, critique can become the same manipulation it seeks to expose.
Circulation outside the film creates another risk. A clip can travel on social media without title, warning or context and become exactly the falsehood the documentary intended to analyse. Productions need to plan for this second life through persistent visual signals and metadata that survive cropping. Meaning is no longer controlled only within a film, but also in the fragments shared beyond it.
The experiment ultimately forces documentary to admit what it has always done: construct a form from reality. AI makes that construction more visible and more dangerous. It can represent absence, visualise contradiction or stage a public figure through their own language; it cannot claim neutrality. Documentary truth is not perfect resemblance, but a verifiable relationship among sources, editing, intention and responsibility.
The experiment shows that a deepfake is not automatically a lie. It can become quotation, satire or analysis. But its truth does not come from photorealism. It comes from the honesty of the film’s method and the viewer’s ability to distinguish document, interpretation and invention. In documentary, synthetic truth remains possible only when the synthesis itself stays visible.