Where it saved us weeks
Impossible worlds. Environments that would have needed a location scout, a permit and a plane ticket now take an afternoon of prompting and a day of cleanup. Backgrounds, skies, dreamlike interiors, the shot that exists for four seconds and would have eaten a third of the budget — that is where the tool pays for itself.
Previsualisation too. We can show an artist the mood of a scene before a single light is rigged, which kills the most expensive kind of argument: the one that happens on set.
Where it broke
- Faces over time — the artist has to still be the artist in frame 200. Identity drifts, and audiences notice instantly.
- Hands and instruments — the fretboard is a lie detector. Musicians spot wrong fingering in a second.
- Continuity — the jacket changes. Every time.
- Lip sync at close range — passable in a wide, uncanny in a close-up.
We integrate generative AI to build impossible worlds — but the artist is still shot on a real camera, by real people, under real lights.
The pipeline that actually works
Shoot the performance properly. ARRI, RED or Sony, on a lit set, with chroma key where we need it. Then build the impossible around it — AI for worlds, 3D for characters, VFX for the seams. The human stays human; the universe gets strange.
That split is the whole method. Use the tool where it is genuinely faster than a crew, and never where an audience can tell.