Can AI make a film with soul?
Updated October 1 2026
What if your life was only ever a prompt?
Typed. Rendered. Handed back to you as video.
Could it catch the rush of a late-night edit? The moment a scene finally clicks. The quiet shock when a stranger on screen says the thing you've never said out loud, yet somehow recognise.
And if it could look right. If it could sound right. If it could make you feel something real.
What would be left that was unmistakably yours?
My short answer is not on its own. AI video can now look and sound like cinema for long enough to be believed. But the thing audiences call soul comes from choices made under pressure by someone with something at stake. A model can produce the image. It can't want anything from it. Whether a film made with AI has soul depends on whether a person is still doing the choosing, and what kind of choosing they're doing.
Steven Spielberg put it more bluntly this year, saying he doesn't believe anything can substitute for the soul. I agree with him. But soul can sound like a mystical word for something quite practical, and if we can't say what it is, we won't notice when we're losing it.
What soul actually is in a film
We learn as kids that 1 + 1 equals 2, a measurable truth. Grown-up culture loves the same comfort. Mensa scores, rankings, neat proxies for brilliance.
But a gut-wrenching film or Van Gogh's Starry Night doesn't land because it follows a formula. It lands because it carries judgement under pressure, taste, restraint, and the lived context behind a choice. You can analyse it afterwards, but you can't reduce it to a score without losing the point.
Creativity is a chain of gambles you make in real time. Gritty or glossy. Safe or bold. On an indie shoot, a dropped light can throw an unplanned shadow across a face and suddenly the scene changes. You either notice it and lean in, or you miss it and keep going.
That responsiveness isn't mystical. It's attention meeting circumstance.
Picture a pianist wrestling Beethoven. Fingers stumble, a chord wavers, yet the room still feels the pulse. Imperfect struggle moves us because it proves there were stakes.
Flawless AI Chopin can feel like supermarket Muzak with a doctorate. Not because it's bad, but because the waver is the bit that whispers someone risked something.
If AI can generate flawless Chopin, the question is whether flawless is what you were listening for.
Human passion and machine precision in the same room.
What AI video can do now
The shift isn't that AI can make a moving image. It's that the image is starting to hold up for longer, with more control and fewer obvious tells.
When Google released Veo 3.1 in January 2026, it added native vertical video, upscaling to 1080p and 4K, and a stronger way of turning reference images into scenes that hold together. The rival models have been racing in the same direction, towards consistent characters, steadier camera moves and longer takes.
That's the line you feel in your stomach. When a generated clip can sustain belief, it stops behaving like a gimmick and starts behaving like footage.
You don't need to memorise model names to sense what's happening. Short clips are easy to dismiss. Scenes are harder. Scenes create memories. Scenes create belief.
Left, the raw joy of human creativity. Right, a step into an unknown future shaped by AI. Which direction does cinema choose next.
Where the choosing goes
On a shoot, the accidents arrive whether you want them or not. The dropped light. The stranger who walks through the back of frame. The interviewee who says the true thing just after you've called cut. Your job is to notice, and to decide in the moment whether it's a mistake or a gift.
With a prompt, the accidents still happen. They just happen in the render queue, as twenty versions of the same shot, each slightly wrong in a different way. The model doesn't know which one matters. Somebody still has to.
So the soul question moves. It isn't in the camera and it isn't in the model. It's in the person deciding which version is true, and whether they're willing to throw away the one that's merely impressive.
Choosing from twenty outputs is a real decision, but it's a different kind from catching a shadow as it falls across a face. One is selection. The other is response. A film built only from selection tends to feel like a catalogue, polished and strangely weightless, because nothing in it ever happened to anyone.
The real risk is the average
There's a creative risk too, and it looks polite at first. When the easiest path becomes remixing existing styles on demand, the average smooths out. Not because people want bland work, but because defaults are powerful and speed rewards sameness.
A lot of films already borrow their emotional shape from older films. AI doesn't invent that. It industrialises it.
Filmmaking also isn't one creator and a laptop. It's a network of people and specialist crafts that exists because production is hard and unpredictable. If automation thins out that network, it doesn't only change budgets. It changes who gets to learn the craft in the first place.
The danger isn't only that machines can make images. It's that speed rewards us for not trying, and we start calling the average good enough because it arrives on time.
When anyone can be generated
The obvious risk is the deepfake. The quieter one is the default shrug, where any clip that turns out to be inconvenient can be waved away as probably AI.
Picture this. A clip of you, your face, your voice, saying something you'd never say. It spreads while you sleep. By morning your phone is hot in your hand, and you already know before you open it that something has shifted. Someone calls, not to ask how you are, but to ask if it's true.
You say no. Then there's that pause. Not long. Just long enough.
The damage isn't only the lie. It's the wait.
Now imagine it isn't public. It's emailed to HR, or dropped into a family group chat with no context. There's no investigation first. There's only mood.
Consent sits right in the middle of this. Performers can see where "just for this production" quietly turns into "forever, everywhere", which is why Equity members voted to refuse digital scanning in December 2025. On the production side, getting proper consent for a digital replica is now a job in its own right.
Once doubt is cheap, where something came from starts to matter more than how it looks. Content Credentials from the Coalition for Content Provenance and Authenticity record a file's history, so people can inspect how it was made and changed. It won't fix everything, and a clip can lose its credentials the moment it's cropped and reposted without its context. But it gives honest makers something stronger than trust me.
Next time a clip guts you, ask yourself three things. Who wins if you swallow it whole? What proof would actually move you? And if none comes, why are you still believing? There's more on how audiences decide what to trust if you want to go further.
If the copy can outshine you
You don't have to be anti-tech to feel the tension. You can love new tools and still care about what they do to belief.
Cinema isn't only results. Process shapes voice. The messy part, where you're not sure the story lands yet, is often where you find out what you actually think.
If everything can be generated quickly, the temptation is to skip that uncertainty. That's not only a creative loss. It's a trust loss too, because the audience has fewer signals that something was chosen rather than merely produced.
Just because we can flood the world with flawless counterfeit lives, yours included, should we?
At some point, someone will prompt a version of you. It might not even be malicious. A joke, a dare, a marketing test, a bored stranger.
The version they get back will look more alive than the one reading this. More convincing. Easier to love.
How will you prove the original was ever real?
Questions people ask
Can AI make a film with soul?
Not on its own. AI can now produce footage that looks and sounds cinematic, but what audiences feel as soul comes from human choices made under pressure. A film made with AI can still have it, as long as a person is doing the deciding and is willing to reject the result that's merely impressive.
Why does AI video often feel soulless?
Usually because nothing in it happened to anyone. Generated footage leans towards the average of what has worked before, and it's built by selecting from outputs rather than responding to a moment. When a person makes real decisions with it, including throwing away the impressive version, it can carry intent. Left on its defaults, it rarely does.
Is choosing between AI outputs real filmmaking?
It's a real creative decision, but a different one. Selecting the best of twenty generated versions is curation. Responding to something that happens in front of a camera is direction. Most good work made with AI will need both, and the films that feel hollow are usually the ones built from selection alone.
