A launch film I could direct in words and render exactly again.
The Idea
When VertX was finished, I had a real launch film to make and a question I wanted to explore: what might motion design look like when an LLM becomes part of the production process? As a senior motion designer, I wanted to test that possibility without giving up the control that makes the work mine.
Editability was non-negotiable. I wanted to change a line, a colour, a movement or a cue and know that the next render would reflect that decision exactly. That ruled out generating video frames as a one-off result. Instead, I set out to direct an LLM through the creation of a motion system I could inspect, revise and render as often as I needed.
How It Works
- I gave Claude Code a factual product brief. We developed the film's script and visual sequence, then established design and motion guidelines to keep the work coherent.
- Claude built browser prototypes that I could watch and respond to. I chose the exploded isometric approach, then directed revisions to the artwork, copy, movement and pacing.
- Those scenes were drawn in JavaScript on a 2D Canvas. Each frame depends on its position in time, so I could scrub to any moment and render it again.
- We brought the scenes together in an animatic, worked out the transitions and placed motion cues against the music. Feedback stayed concrete: what felt too busy, what needed more time and what changed the meaning of a shot.
- React and Remotion supplied the frame clock, video render and audio. The visuals came from the same scene code I had reviewed in the browser, with checks for unwanted jumps before the final render.
What I Learned
The process felt familiar in one respect. As in After Effects, Blender or Cinema 4D, we moved from an idea through a visual plan, rough versions, an edit and polish. What changed was my position in the work. I spent less time constructing each moving element myself and more time directing what Claude built, reviewing it and asking for specific revisions.
I supplied the real app capture and icon, chose the music and provided the fonts. The remaining visuals and motion were built in code through that back-and-forth. The result is AI-assisted, not a black-box AI-generated video: its scenes remain editable, its timing can be changed and the same code and assets render the same frames every time.
