Problem
Set design runs at the speed of iteration: an idea has to be built or drawn before anyone can judge it. Filmprovising puts the rapid iteration AI makes possible behind the designer’s own blocks and presets, so they can try many rooms quickly while the layout and the calls stay theirs.
The interface
123456
Designer
- 1Picks a saved scene and shot on the slate.
- 2Gives a line of direction.
- 3Sets Control: how far the render may stray from the blocks.
- 4Captures the frame.
Model
- 5Renders the camera frame as a finished set, keeping the layout of the blocks.
- 6Make into video sends the chosen render to Runway.
System
Why both
Exploring runs locally: each render costs nothing, the loop stays fast, and unreleased designs never leave the room. The API comes in when quality matters more than count, and Runway only for the few frames worth turning into motion.
The models changed under it every few months, so the model is the swappable part: the camera, the presets and the interface stay put while it moves from one release to the next.
Beta
I tested the beta with 10 set designers and 10 filmmakers, asking how they would use it in their own work. What they liked: rapid iteration, and scene elements they could build once and keep across many generation rounds. The beta asked how they would use it, not how fast it got them there; next I would measure how many rounds it takes a designer to reach a set they would take to the build.