- What
- A short presenter film, fronted by a photoreal AI version of me
- Running time
- 63 seconds
- Performance
- Mine, recorded in one take on a laptop webcam, reading from an autocue
- Voice
- Mine, cloned from a three-minute studio recording
- Set
- A darkened studio with a cinema camera behind me. It doesn't exist.
- Angles
- Studio wide and an 85 mm close-up, cut on natural pauses
- Turnaround
- Laptop take to finished film in one evening
- AI generation cost
- About US$30 for the finished film, its test and the character sheet
The idea
Leaders get asked to front films all the time, and most of them would rather not. A studio day means lights, make-up, a crew and a diary nobody can clear. So I tried it on myself first. Could I sit at my desk, read my script once into a laptop, and get back a film that looks like it was shot in a proper studio?
AI Dave is the answer. He says who he is in his first line, because a likeness should always be labelled. Everything he does, he does because I did it first.
Why it matters
It's the same approach I'd use for a CEO update, a training series or a product launch: one relaxed take from wherever the speaker happens to be, turned into a consistent, on-brand studio look, with the speaker's own face and voice, and their sign-off on every frame and every word.
My role
Writer, performer, director and editor. I wrote the script, designed the look, gave the performance and cut the film. The AI handled the studio, the lighting and the camera.
How it was made
01
Casting myself
AI Dave was cast like any other character: one locked identity sheet with a full-length turnaround and four expressions, built from my reference photos. From that face came two studio set-ups, a wide and an 85 mm close-up, lit the way I'd light a real interview: three-point light, a darkened studio, a deep teal background and a cinema camera out of focus behind me.
02
One take, on a laptop
My first attempt let the AI perform the script by itself. It looked fine, but it fluffed words, and it wasn't me. So I threw it away and gave the performance myself: one take at home, in a T-shirt, on a laptop webcam, reading from a scrolling autocue.
Motion transfer then carries every head turn, gesture, pause and breath from my take onto the studio version of me. The ten-second test on the right is the moment we knew it worked.
03
My voice, minus the laptop
Laptop audio sounds like a laptop. So the sound was swapped for a clone of my own voice, made from a clean studio recording. Because it follows my original take, every word and pause stays exactly where I put it.
04
Two angles, cut on the pauses
Motion transfer works in clips of up to 30 seconds, so I split my take where I naturally paused and cut between the wide and the 85 mm close-up, just as I would with a two-camera interview.
05
The edit
Assembled, levelled for broadcast loudness and handed over as a Final Cut Pro project with every clip, so the cut points can still be moved by hand.
Toolset. Stills: Nano Banana Pro. Performance: motion transfer from my own laptop take. Voice: a clone of my own voice. Edit: Final Cut Pro. Orchestrated through an agentic AI workflow that handled prompts, autocue, version control and cost tracking, so I could stay on the creative decisions.
What I'd do differently
Honestly? I'm not thrilled. For one thing, my wife now prefers AI Dave. More seriously, the cut to the close-up halfway through lands like a jump scare: a side angle or a looser frame would have been kinder. It's also only 720p, so the picture is a little soft. All fixable, but for a proof of concept I couldn't justify another US$30 on a retake.
Why it matters for your next project
- No studio day.
- Your speaker records one relaxed take wherever they are. The studio, the lighting and the camera come later.
- Their face, their voice, their words.
- The performance and the voice are the speaker's own, and the script is kept word for word.
- Consent and sign-off, always.
- Nobody's likeness is used without their written consent, and they approve every frame and every word.
- Honest by design.
- AI presenters are labelled as AI. AI Dave says so in his first line.
By the numbers
- 63 s
- Running time
- 1
- Take, recorded on a laptop
- 2
- Angles: studio wide and 85 mm
- 3
- Clips, cut together on natural pauses
- 0
- Studio, crew or lighting rig
- 1 evening
- Laptop take to finished film
- ~US$30
- AI generation cost of the finished film
- 1
- Human in the loop