Where AI is used in film and TV production

AI is already used across film and TV production, but most current use is less dramatic than the public debate suggests. The clearest examples sit in development, pre-production, post-production, localisation and marketing: research support, script breakdowns, footage search, cleanup, dubbing and the production of promotional variants. Some AI-generated material does reach the screen, and practice varies widely between companies and productions, but in many workflows the tool is assisting a task rather than making the final creative decision.

Available evidence remains partial, so this is better understood as a snapshot than an industry census. The CNC’s AI observatory (opens in a new tab) tracks adoption across film, television, animation, post-production and VFX, while a January 2026 industry report (opens in a new tab) found early use concentrated in development, pre-production, localisation and selected post-production workflows. Both point to the same broad picture: adoption is real, but uneven, and the quieter applications remain more common than the headline-grabbing ones.

Development and pre-production

In development, generative tools turn up as research and drafting aids. They can produce first-pass coverage, sketch loglines, suggest comparable titles and act as a brainstorming partner that never tires of a bad idea, which is useful precisely because the output is disposable.

That is different from saying AI is writing the films that get made. On projects covered by the Writers Guild of America, AI-generated writing is not treated as literary material, and a writer may choose to use AI only with the company’s consent and subject to its policies. Elsewhere, including across Europe, the contractual position varies. In either case, “AI-assisted writing” is too broad a label to tell you who contributed what, which is one reason the individual use needs to be recorded more precisely.

Some of the most practical applications appear in preparation. AI can help turn a script into a first-pass breakdown, organise scheduling and budgeting information, or produce early storyboard and previs material. Industry interviews have reported modest productivity gains in selected workflows, although those findings should not be mistaken for a measurable saving across an entire production.

Because much of this material is intermediate rather than public-facing, disclosure questions may be less immediate. The risk is not zero: scripts may be confidential, inputs may contain personal data or protected material, and the terms of the tool may restrict what can safely be uploaded.

Post-production

In the edit suite, AI often appears as a set of faster hands rather than a new head. It can transcribe and log rushes, tag metadata, search footage and accelerate parts of rotoscoping and cleanup. These functions reduce the work involved in handling large amounts of material, but the editor still decides what the scene means, where it turns and what should be left out.

The term “AI” also covers several different kinds of technology. Transcription, object tracking and footage search are not the same as generating a new image or performance, and they do not create the same questions. A useful account therefore records the task and the output, not merely that “AI was used in post.”

Localisation and marketing

Localisation is one of the more visible areas of adoption, driven by the economics of releasing in multiple territories. AI-assisted speech synthesis, voice matching and lip synchronisation can shorten parts of the dubbing workflow, but quality remains inconsistent and the choices about language, tone and performance still require human review. Where a recognisable voice or performance is involved, consent, contract and compensation also matter.

Marketing is another natural testing ground because the work depends on variation. Key art in different dimensions, social cutdowns and alternate trailers for different territories all create a high volume of deadline-bound work, and generative tools can help produce drafts and adaptations. This is also one of the places where AI-generated or manipulated material can reach the public, making it important to know how the asset was made and where it appeared.

Why the production record matters

The difficulty is not that every individual use is alarming. It is that the uses are spread across departments and vendors, while the record, if one exists, is often fragmented.

Copyright and authorship questions are easier to examine when there is a contemporaneous account of who did what and with which tools, rather than a reconstruction made after delivery. The same applies when an insurer, broadcaster or commissioner asks about AI use as part of its own due diligence.

The EU AI Act adds a more specific reason to know what happened. Since 2 August 2026, Article 50 (opens in a new tab) has required organisations using an AI system to disclose generated or manipulated image, audio or video when the content meets the Act’s definition of a deepfake. For evidently artistic, creative or fictional works, the disclosure can be made in a way that does not interfere with the experience of the work.

This does not mean that every AI-assisted asset must be labelled. It does mean that a producer cannot assess the obligation without knowing what was generated or manipulated, what it depicts and where it was used.

A useful starting point

The response can be modest. A music cue sheet offers a useful analogy, though not an exact legal equivalent: keep a production-level record of the tools used, the tasks they supported and whether their outputs entered the finished work or its promotion.

A record like this does not prove what happened in the way a forensic system would, and it does not establish compliance by itself. What it preserves is a dated account of what the people involved reported, which puts a producer in a far stronger position than trying to reconstruct the production months after delivery.