AI in filmmaking in 2026: what has actually changed, and what hasn't
Sora is gone, Veo, Kling and Runway have split the market, and most marketers now use AI somewhere in their video workflow. A filmmaker's honest view of where AI helps, where it hurts, and how founders should use it.
Every few months someone tells me AI is about to make filmmakers obsolete. Then I watch the same people spend a fortune trying to get a generated face to say one sentence convincingly.
I run a film studio, and we use AI every day. So here is the honest version: what has genuinely changed in 2026, what hasn't, and how I'd advise a founder or speaker to use it.
What changed in AI video this year?
The biggest headline is a departure. OpenAI shut down Sora. It announced the decision on 24 March 2026, closed the consumer app and website in April, and scheduled the API to follow in September. Press reports linked the move to the enormous computing cost of video generation and a shift in priorities towards coding and enterprise tools.
That left a market with three clear leaders, each good at a different job:
- Google Veo (3.1) — the most reliable all-rounder. Strong prompt adherence, native audio, high resolution. Google watermarks the output with SynthID.
- Kling (3.0) — Kuaishou's model moved from "budget option" to the top tier, with longer shots and strong multi-shot sequences.
- Runway (Gen-4.5 and Aleph) — positioned for professionals. Its standout trick isn't generating from nothing; it's editing what you already shot: changing the angle, the lighting, an object or the whole look of a real clip.
The second change is quieter but more important: AI is now normal in marketing teams. Wyzowl's 2026 research, as reported by several outlets, found that around 63% of video marketers have used AI tools to create or edit videos. If you're wondering whether to use it at all, the market has already answered.
What hasn't changed?
Three things, and they matter more than the tools.
Clips are still short. Most models top out at seconds, not minutes. Real projects still mean stitching shots together, which is where most of the craft — and most of the cost — still lives.
Consistency is still hard. Keeping the same face, wardrobe and light across twenty shots is something a camera does for free and a model fights you on.
Trust is still human. This is the one that founders forget. When you're building a personal brand, the product is you — your face, your voice, your judgement. An audience can't build trust with something they suspect isn't real. And audiences are getting very good at suspecting.
Where AI genuinely earns its place
In our studio, AI is a power tool, not the talent. Here is where it pays for itself:
- Concepts and pre-visualisation. Before a shoot we can show a client what a scene will feel like. It's faster than a mood board and far cheaper than finding out on the day.
- B-roll and cutaways. The establishing shot of a city at dawn, the abstract visual behind a statistic, the product detail you forgot to capture. Generated B-roll — clearly not pretending to be documentary — is a great fit.
- Clean-up and fixes. Removing a boom mic, extending a background, fixing a distracting object. Hours of rotoscoping become minutes.
- Versions. One hero cut becomes a vertical, a square, a 15-second ad and a translated version. AI makes the variations; humans decide which one goes where.
- Scripting support. Drafting hooks, finding angles, structuring a story. Then a human rewrites it until it sounds like a person.
Where it backfires
- Fake founders. An AI avatar delivering your "personal" message is the fastest way to tell your audience you didn't think they were worth your time.
- Synthetic testimonials and "customers". Beyond the trust damage, this is the kind of thing that gets brands into regulatory trouble.
- Undisclosed realism. If a shot could be mistaken for real footage of a real event, label it. You lose nothing by being honest and you risk everything by not being.
How I'd advise a founder to use AI in 2026
Keep the human core real and let AI multiply it:
- Film yourself properly — one well-planned shoot day a month gives you the raw material.
- Use AI around that footage — B-roll, versions, captions, translations, clean-up.
- Use AI to think faster — angles, hooks and structures, which you then make your own.
- Disclose anything synthetic that could be mistaken for real.
Generating a shot is now cheap. Knowing which shot to make, who it's for, and how it turns attention into revenue is not. That part was never really about the camera, and it isn't about the model either.
Frequently asked questions
Is Sora still available in 2026?
No. OpenAI announced on 24 March 2026 that it would discontinue Sora. The consumer app and website closed in April 2026, with the API scheduled to follow in September 2026.
What is the best AI video generator in 2026?
There is no single winner. Google's Veo 3.1 is widely seen as the safest all-rounder with native audio, Kling 3.0 is strong for multi-shot sequences and value, and Runway focuses on professional editing tools such as changing the angle, lighting or style of existing footage.
Should founders use AI avatars instead of filming themselves?
For most founders, no. People buy from people they trust, and audiences are getting better at spotting synthetic faces. Use AI around your real footage — for B-roll, versions, translations and clean-up — rather than as a replacement for you.
Will AI replace video production companies?
It is replacing some tasks, not the job. Generating a shot is now cheap. Knowing which shot to make, why, for whom, and how it fits a strategy that turns attention into revenue is still human work.
