Chris Hogg, who lectures Creative and Social Media at Royal Holloway, invited me to come and run a workshop with his multimedia students on using generative AI for creative storytelling. Also in the room was Professor David Howard, who’s known for recreating the voice of Nesyamun, a 3,000-year-old mummified Egyptian priest, by 3D-printing his vocal tract from a CT scan and pushing air through it. A single short aaah sound, but a real one. That detail matters later.
I called the talk “Cinematic Synthesis: Generative AI in Creative Storytelling”. The plan was an hour of how AI image generation, voice cloning, and 3D content creation actually work, then a four-hour build: each team makes a short movie trailer for any film they want, using only generative tools. Whole thing in one afternoon.
The tools
Started with Midjourney. Even compared to a year before, the leap in image quality was hard to absorb in real time. We looked at prompt structure, the way small changes (lens specification, lighting, art-movement reference) completely change what comes back, and the difference between describing a shot and describing a scene.
Then we got into 3D and motion. Luma NeRFs for turning a phone-shot scene into a navigable 3D capture. Move.ai for markerless motion capture from regular video. Wonder Dynamics for swapping a live actor for a CG character in existing footage. We also looked at AI video generation, which in May 2023 was still a hot mess most of the time, but with the kind of edge moments that made it clear where it was going.
The point I tried to make wasn’t here are the tools. It was here is how cheaply you can now skip the production-quality bottleneck if you have a good idea. Storyboarding to camera-ready stills in minutes. Mocap without a suit. Voice cloning from a few seconds of reference. Each of these was a full production-house budget line a year before.
The workshop
The students split into teams and had four hours to make trailers for whatever film they wanted. I floated between groups answering tool questions and using ChatGPT alongside them to generate ideas, plot outlines, shotlists, working dialogue, sometimes whole scene treatments when a group hit a wall.
Four hours is not long for a film trailer. What came out was, candidly, beyond what I expected. None of these students had touched any of these tools before that morning. By 4 p.m. several of them had things you could put in front of a tutor for a storyboard mark and a few that would have been good enough for a real pitch reel.
What stuck with me wasn’t the polish. It was watching the ideas un-cap. Half a dozen times that afternoon I heard a variant of I always wanted to make this but I couldn’t. Big concept ideas they’d thought were out of reach because they didn’t have a stunt budget, or a CGI house, or twelve takes with an actor who’d already been booked elsewhere. The tools removed those specific constraints for an afternoon and the ideas that came back through were genuinely ambitious. Some students went straight for cinematic restraint; others built absurdist pitches the moment the budget excuse dropped out from under them.
The mummy
The professors made their own team. Obviously their film was about the mummy.
Chris and David ended up making a horror trailer in which Nesyamun returns to reclaim his voice. Genuinely unsettling. The framing was particularly good, because the premise was a real research thread: David had spent years exploring what AI might do with the limited viral aaah of the original recording (could you push a vocal tract organ + a generative voice model far enough to give a mummified priest full dialogue?). The trailer was, in a sense, the speculative version of his own future paper, played for horror. The most academically rigorous people in the room produced the creepiest thing of the day, and I’m still not sure that’s an accident.
Takeaway
The main thing I took from it was the speed. Students went from zero to making films in an afternoon, fluently enough that several of them carried on the next day. The barrier to making this kind of content is basically gone now. What’s still hard, and what these tools don’t compress, is knowing what to make and why it matters. That’s the bit nobody can hand you in a four-hour workshop.
The other thing, which I keep coming back to: when an artist with a real research thread gets these tools, the tools become a sketchpad for ideas they’ve been carrying around for years. David didn’t need generative AI to think about the mummy’s voice; he was already thinking about it. The tools just let him show what the thought looked like, in an afternoon, in a horror trailer made with a journalist’s colleague at a university workshop.
Cheers to Chris for having me, the students for getting stuck in, and the professors for the mummy content.