AI in video production: what it speeds up, what it can't replace

Buyers are asking. In the spring 2026 survey by GWA, the German agency association, AI competence topped the list of rising client demands, named by 61.1% of agencies, and 49% rated AI replacing their services as a significant growth barrier (GWA Frühjahrsmonitor 2026 (opens in a new tab)). Here's the honest split.

Where does AI actually save time in video production?

In the hours between the shoot and the first cut.

A 40-minute interview on two cameras is 80 minutes of footage before a single b-roll shot. Someone has to watch all of it, note what was said, find the clean take of every answer and remember where the good close-ups are. That's assistant-editor work, and it's where today's tools are genuinely useful.

What can't AI replace?

Anything that needs judgement, trust, or someone to answer for it.

• The brief. Who is the film for, and what should it change? A tool can format a brief. It can't sit with your sales team and find out why deals stall (what a good brief contains).
• The interview. People open up to a person, not a prompt. Getting a nervous engineer to say the important thing in one clear sentence is a skill (more on interview films).
• Directing people on camera. Where to stand, when to go again, and when to stop because the first take was the honest one.
• Taste and story. AI can find the cleanest take. Choosing the true one, and deciding what to cut entirely, is where a film gets made.
• The final edit and the accountability. When a film goes out under your brand, a person signs it off and answers for every frame.

Our ADAM Audio "In The Studio" series is a good test case: portraits of producers, engineers and artists, shot in London, Berlin and Vienna, nearly always on two cameras, and on three when two people were interviewed together. So every answer exists at least twice. AI can take on the sorting. What makes an episode work is the conversation in the studio, and the decision about which moments carry the portrait.

Is my video AI-generated?

If it was filmed and nothing synthetic was added, no. AI may have helped sort, clean up or caption real footage. That's AI-assisted, not AI-generated.

There are three levels:

1. AI-assisted. Real footage. AI helped with transcripts, logging, noise reduction, colour or captions.
2. AI-generated elements. A generated background, a synthetic voice-over, a generated shot inside an otherwise filmed piece. These should be agreed in writing, and some need a label.
3. Synthetic. An AI avatar of your CEO, a cloned voice, realistic footage of something that never happened. This is what the law calls a deepfake.

Ask your studio which level applies. A good one answers in a sentence; get that sentence in writing. When our World Warfare & Economics release trailer needed a voice, we cast a voice actor, and a film composer wrote the original score.

What does the EU AI Act require since 2 August 2026?

Label deepfakes: realistic synthetic people, cloned voices, fake footage that looks real. Normal AI-assisted editing needs no label.

Article 50 of the EU AI Act has applied since 2 August 2026 (opens in a new tab). The AI Omnibus (opens in a new tab) of July 2026 pushed back other deadlines, not this one. The only grace period, to 2 December 2026, is for the machine-readable marking built into AI tools already on the market (Commission FAQ (opens in a new tab)).

• What counts. A deepfake is AI-generated or manipulated image, audio or video that resembles existing people, objects, places or events and would falsely appear authentic (Art. 3(60) (opens in a new tab)).
• Who labels. The "deployer": whoever uses the AI system under their own authority (Art. 50(4) (opens in a new tab)). Under the Commission's guidelines (opens in a new tab), a company that simply commissions an agency, without deciding whether and how it uses AI, isn't the deployer. Ask for the avatar yourself and that can change. Either way, agree in writing who labels what, and how.
• When. Clearly, and at the latest when someone first sees or hears the content (Art. 50(5) (opens in a new tab)).
• Creative work. For evidently artistic, creative, satirical or fictional work, the label may be done in a way that doesn't spoil it. The Commission reads "evidently" strictly (opens in a new tab): an ad may qualify in some situations, not in others.
• Not retroactive. Content made before 2 August 2026 doesn't need a label after the fact (Commission FAQ (opens in a new tab)).

Based on the Commission's Article 50 guidelines (opens in a new tab).

Where does my footage go when a studio uses AI?

It depends on the tool, so ask. Some AI runs on the editor's computer; a lot runs in someone else's cloud.

With a cloud tool, your files, or material made from them such as transcripts and stills, go to that provider under its terms. If you're an agency under NDA, or a brand filming an unreleased product, settle this before the shoot. Ask:

1. Which AI tools will touch our footage, audio, transcripts or scripts?
2. Does each run on your machines or in a cloud? Which provider, which region?
3. What exactly gets uploaded: full files, audio, transcripts, stills?
4. Can our material be used to train models? Is that switched off, and on which plan?
5. How long is it kept, and how is it deleted?
6. Does any tool recognise faces, and where does that run?
7. Is there a data processing agreement with each provider?
8. Can you run our project without cloud AI if we ask, and is that in the contract or NDA?

What is CutterFly, and why is a film studio building software?

CutterFly is a Mac app for the work before the first cut. Our founder, Sunny Bar, built it, and it isn't out yet.

The problem it's built around, in its own words (opens in a new tab): "The hard part isn't the cut. It's everything before it."

CutterFly, "the AI first-cut editor for Mac" (opens in a new tab), runs on Apple Silicon and studies footage like an assistant editor: it transcribes the audio, groups retakes by topic and marks the best delivery, and describes every b-roll moment so shots can land on the words they illustrate.

Then you brief it, and it builds a cut you refine by hand. The cut goes to Premiere Pro or DaVinci Resolve as an editable timeline, so an editor still makes the final decisions.

CutterFly can run fully locally on the Mac, or analyse footage in CutterFly's cloud. So the questions above apply to CutterFly too.

We haven't used CutterFly on client projects yet; we're about to start. If we archive a client's footage, we keep its CutterFly analysis with it. You can join the early-access wishlist at cutterfly.ai (opens in a new tab).

How we approach this

We choose AI tools project by project. Generative AI goes into your film only if you ask for it.

AI doesn't change our four steps: a real conversation first, then a brief everyone signs off before any camera rolls. We use generative AI, meaning generated video, images, voice-over, music or graphics, only when the client asks for it. We have never used it in a client film without that request. If your project is under NDA, bring the questions above to the first call.

This is general information, not legal advice.

Want to know exactly where AI would and wouldn't touch your film? Ask us on a first call. Let's talk