Cinema put to the AI test: who would fund a film if they didn’t know how it was made?

The problem, however, remains, and clearer guidance from the Ministry will be needed, not least because – as many industry professionals have pointed out – the boundaries are difficult to define. AI can correct a take, extend a performance or replace it. The implications for artists’ work and rights vary considerably. Remuneration arrangements should also be able to distinguish between these uses without forcing every production to renegotiate the meaning of the terms from scratch.

Giacone highlighted the industrial and employment-related role of public support, stating that the ministry is committed to identifying criteria for assessing new technological costs. This is where the issue becomes political. What does the State wish to fund? The projects themselves, the labour required to carry them out, the development of skills, or businesses’ ability to compete? These are objectives that can coexist, but AI brings the tensions into sharper focus. A manufacturer’s cost saving may correspond to a lost contract for a supplier. Conversely, an overly rigid rule may discourage the use of tools that would enable a filmmaker to shoot a scene that would otherwise be beyond budget.

According to lawyer Chiara Dellacasa, it is when drafting contracts that the issues really come to a head: modifying a performance and creating a digital replica require different agreements. The MiC’s interpretative guidelines stipulate, amongst other things, explicit consent and remuneration commensurate with the personal contribution for the use of digital replicas or synthetic performers. Modifying or supplementing an author’s work using generative AI requires separate written consent.

It remains to be seen to what extent the producer can control what happens between the signing of a contract and the delivery of the material. A contributor uploads an image to a free service. A supplier uses a generative function during routine processing. Someone tries out a programme to speed up a delivery. Each step may introduce different terms of use: what commercial uses are permitted? Are the materials stored? Can they be reused to train a model? A contractual guarantee is necessary, but it is not enough to reconstruct what has happened.

Tech companies such as that of Siliva Paonessa, CEO of CrewReel – the innovative start-up that organised the event – aspire to play a role in this. Paonessa proposes documenting and tracking the production chain through a system that collects information on software, licences and roles, alongside evidence of human decisions and versions of materials. File digital fingerprints, timestamps and signatures help to make the file verifiable. The aim, she explains, is verification within the production process. When looking at the finished film, determining a precise percentage of human creativity means asking software for an answer that also depends on choices, authorisations and responsibilities. Knowing who decided on a particular change, using which materials and with what consent, provides more useful information than a label added at the last minute.

However, traceability also comes at a cost. If documenting the use of a programme requires an administrative structure accessible only to large organisations, smaller production units are at a disadvantage from the outset. We need proportionate procedures and agreed criteria: precise enough to allow for checks, yet practical enough to be applied whilst work is in progress.

Further complicating matters are the initial findings emerging from the convergence of AI and cinema. In China, the first ambitious science-fiction film shot entirely using AI – entitled *Sanxingdui: Future Past* – appears to have grossed very little, despite a budget of around 1.5 million. Similarly, the first episode of the new *Odyssey* series, created using Grok, does not seem to have won over audiences. It is not, therefore, a foregone conclusion that AI will kill off art and culture. However, understanding how a cultural product is created could become important – not so much to determine its success, but to understand how (human) talent should be funded.