When anyone can generate anything, the work is knowing what to keep.
I spent 25 years building worlds on real film and commercial sets. Now I make films with AI, and I bring the same discipline: references, decisions, corrections, and a bar that does not move.
-
01
AI is a craft tool, not a shortcut
I treat generative AI the way I treated a set: it needs art direction. Every image starts from a reference, a decision and a reason. The bar is simple and strict: the result has to hold up next to real production.
-
02
Judgement over tools
Anyone can generate an image. The work is choosing. Every film on this site shows where it started, what I tried and what I threw away. The discards are not hidden: they are the clearest proof of judgement.
See the discards in MUSES → -
03
A method a team can use
For years I ran art departments of up to 50 people toward a single look: planning, hierarchy, control. I run AI pipelines the same way, step by step and documented, so the work never depends on one person or one lucky prompt.
-
04
Where I don’t use AI
People first. When I can film real people, I do. A real choir, a real performance, a real face in front of a camera beats anything I can generate, at least for now. AI comes in where the camera can’t reach.
I test before I trust. I have tried many things with AI that did not work, and I say so. Crowds are a clear case: today’s tools still struggle with them, and you can see it in my early experiments. I keep what holds up, discard what doesn’t, and show both.
Where the tools fail ↓ -
05
Likeness, disclosure, provenance
Likeness. The characters in my own work are invented. None is modelled on a real person, and I don’t use anyone’s face or voice without their consent.
Disclosure. Every piece says what was filmed and what was generated, and comes with its making of, so you can see how it was built.
Provenance. I draw references from film, photography and art, as any art director does, to define look and mood. What is generated is my own work, made from my own prompts and decisions. The making of shows where each piece comes from, including what I tried and discarded.
Tributes. Two early experiments, Echoes of Pink Floyd and Smoking Stones, were made for tribute-band concerts and quote those bands’ visual world. I label them as tributes and show them as experiments, not as client work.
Where the tools fail (so far)
Every discard on this site has a reason. These are the failures I run into most, in real clips from my own films.
The audience fills in, but the faces and bodies don’t hold.
Echoes of Pink Floyd · 2024Asked for the Palau, it built a generic opera house. The hall had to be ours.
Echoes of Pink Floyd · 2024A fast move tears the architecture apart.
Smoking Stones · 2025The horse rears up and ends up as two horses.
MUSES · 2025Stone turns into a pale, fleshy horse, and the framing jumps.
MUSES · 2025Left alone, the light falls back on a stock rock show.
Echoes of Pink Floyd · 2024So real people stay real, the real space gets filmed, and AI animates what the camera can’t move.