We make films with AI. Not “AI-assisted,” not a couple of generated shots dropped into an otherwise normal edit. Complete ads and short films, generated end to end, delivered as the finished piece. We’re good at it, we sell it openly, and we think it’s one of the most exciting things to happen to this craft in years.
We also have one rule we will not break, whatever the tools can do. We will never use AI to make a documentary, or to fake anything a viewer is meant to take as real. That single line is what this piece is about: what we happily generate, what we would never generate, and why the difference has nothing to do with what’s technically possible.
What we make with AI, completely
For the right brief, AI isn’t a helper on the side of the process. It is the process. We’ve built finished ads and short films generated from the ground up and delivered them as the main event, not a novelty reel.
It tends to be the right medium in a few situations:
Ads where the world is invented anyway. A lot of advertising already lives in a heightened, stylised, unreal space. When the concept doesn’t depend on a real location or a real person, generating it can be freer, faster and stranger than a shoot ever could, and the finish can be genuinely beautiful.
Short films with a deliberately unreal aesthetic. If the whole point is a look you couldn’t capture with a camera, AI isn’t a compromise, it’s the instrument. We’ll build the entire thing generatively and treat it with the same rigour as any other film.
Subjects with no physical form. A network, a protocol, a token, a settlement layer. Nobody has footage of these. Generating an image language for them isn’t a substitute for filming something, it’s the honest way to picture something that never existed in front of a lens.
Ideas that would be impossible, unsafe or absurd to shoot. Scenes at a scale no budget could stage, or that simply couldn’t happen in the real world. Generation makes them makeable.
When a piece is generated, we say so, clearly and up front. Labelled as what it is, a fully AI-made ad is a legitimate and exciting form. The problem was never AI. The problem is pretending.
The one thing we will never do: fake real life
Here is the line, and it does not move. If a viewer is meant to believe they’re watching something that actually happened, it has to have actually happened.
So we will never:
- Make a documentary with AI. Real people, real places, real events, captured for real. That’s the entire contract a documentary makes with its audience, and generating it would break it.
- Synthesise a real person’s face, voice or testimony. A founder’s words work because they’re theirs. Manufacturing them doesn’t save time, it removes the only thing that made them worth including.
- Pass generated footage off as reality. No quietly synthetic “b-roll” standing in for something that never occurred, in anything the audience takes as a record of the real world.
The whole power of a documentary is that it’s true. When we travelled to Tanzania to film World Mobile’s Smart Village, the point was precisely that a village really was being connected, and that we were really there when it happened. No model can generate that, because what makes it land isn’t the image. It’s the fact behind the image.
Why the line matters
Audiences are getting fluent in what generated footage looks like, and faster than most brands expect. The moment someone suspects that “real” footage was faked, they don’t just doubt that clip. They doubt everything you’ve ever shown them.
Which is why disclosure isn’t a nicety, it’s the whole game. A labelled AI ad is honest and impressive. An unlabelled synthetic documentary is a liability that no amount of production value can buy back. Keeping generated and real cleanly separated, and clearly signposted, is how you get to use both without poisoning either.
Generated or real, the craft bar is the same
Making a film entirely with AI doesn’t lower the standard, it moves where the work is. The tools get you to a plausible result astonishingly fast, and then the last stretch, the part where a brand’s product has to look like their actual product and a sequence has to actually mean something, is as demanding as it ever was.
That’s because the real constraint was never the kit. It’s taste: knowing what’s worth making, and being able to tell when something is nearly right versus genuinely right. Generation multiplies judgement. Point it at good judgement and it makes you dramatically faster. Point it at none and it makes you faster at producing confident, forgettable work, and there is going to be an enormous amount of that.
How we decide, in practice
Before we reach for a generative tool on client work, we ask three things:
- Is this meant to be taken as real? If yes, we film it. Full stop. If no, generating it is firmly on the table.
- Would the client be comfortable with us labelling exactly how it was made? If the honest label needs a caveat, we’re on the wrong side of the line.
- Is AI the right medium here, or just the fast one? Fast is a perfectly good reason. It just has to be a chosen reason, not a drift.
The point
We’re not AI sceptics and we’re not AI evangelists. We’ll generate an entire ad without blinking, and we’ll get on a plane to film a documentary because no model can make something true. The tool is remarkable. The judgement about when to use it, and when not to, is the whole job.