Less than three weeks remain until August 2, 2026, when the transparency obligations in Article 50 of the European AI Act become applicable in the European Union. For cinema, advertising, music videos, social content and post-production, this is not an abstract issue: it concerns how a synthetic image is produced, recognized and disclosed to the public. The most important point is that transparency does not amount to a single universal label. It changes according to who provides the tool, who uses it and what kind of content is created.
Providers of generative systems are required to make artificially generated or manipulated content detectable in a machine-readable format. In practice, a technical signal should travel with the file and remain identifiable. Professional users of those systems, meanwhile, are responsible for informing the public when images, audio or video constitute synthetic content that could appear authentic. These are complementary layers: a technical trace for systems and a comprehensible disclosure for people.
For works that are clearly artistic, creative, satirical or fictional, the regulation introduces a form of balance. Disclosure must still exist, but it can be presented in an appropriate way without obstructing viewing or undermining the experience of the work. This is especially relevant to cinema: it does not require every sequence to be interrupted by an intrusive warning, but it does require the synthetic origin to be knowable through credits, program notes, descriptions or other points suited to the context of distribution.
The word ‘AI’ in the end credits, however, risks being too vague. A model may have been used for a voice, to extend a set, replace a face, create a crowd, translate lip movement or generate an entire shot. A useful disclosure should indicate at least what part was affected, what function the tool performed and what degree of control the production exercised. There is no need to publish a complete production diary; the point is to avoid hiding profoundly different interventions behind one indistinct formula.
The Code of Practice published by the European Commission offers a voluntary path for demonstrating compliance, while European icons may help make disclosures more uniform. The icons are optional and, on their own, do not prove compliance. For an audiovisual production, the strongest approach is therefore layered: technical information retained in the file when available, a visible disclosure at the point of publication and more detailed credits for anyone who wants to understand the process.
From now on, provenance should be treated as part of the production workflow rather than as a sentence added at the last minute. That means recording which tools enter pre-production, shooting, editing, visual effects and sound; keeping versions; checking that exports do not strip metadata; and agreeing with platforms and distributors where disclosure will appear. This record also protects human work because it distinguishes what was generated from what was directed, performed, photographed or composed.
Transparency does not reduce the creative value of AI. It can do the opposite: it shifts attention from suspicion to choice. When viewers know that part of an image is synthetic, they can judge how it was used instead of merely trying to discover whether it is ‘real’. For cinema and video, the August deadline therefore marks the beginning of a new grammar of credits: less mystery about the process, more responsibility for meaning.