Generate three directions that a viewer would recognise as three different videos. Same brief, same message, different visual language, structure and pacing.
Make them different on purpose
Pick three different starting angles, for example:
- Typographic: the words are the picture (kinetic type, editorial layout, masks, scale contrast).
- System / diagram: the idea as a structure that builds (grid, nodes, flows, data, morphs between states).
- Object / product led: one hero object or the real UI, framed and moved like cinematography.
- Others when they fit: documentary-real (footage, captures), illustrative (a drawn character or metaphor), data story.
Test: if two directions could be the same video with a different palette, replace one.
For each direction write
| Field | What to write |
|---|---|
| Name + one-line idea | "Assemble": the product is built in front of you, piece by piece |
| Visual language | Shapes, materials, texture, depth, what is on screen most of the time |
| Palette | 3 to 5 hex values with roles (background, ink, accent, support) |
| Typography | Families, weights, scale, how text enters |
| Structure | Beat list with rough times (hook, development, payoff, end) |
| Pacing | Cuts per 10 s, where it holds, where it accelerates |
| Animation style | Easing character (spring stiffness / damping, overshoot or not), stagger, camera behaviour |
| Transitions | Which ones and what carries across |
| Audio direction | Music mood and tempo, SFX palette, voice or none, where silence goes |
| Signature moment | The one shot people would remember |
| Risks | What could make it look cheap or generic; what it needs that we may not have |
| Effort / engine | Remotion or HTML engine; external assets needed |
Recommend
End with: which direction is strongest for this purpose and audience and why (two or three concrete reasons tied to the brief), what you would borrow from the others, and what the user should decide. Then stop and wait for the pick. Do not storyboard three directions.
Subagent option
Spawn three subagents in one message, one per angle, each given the brief and this file, each returning one direction in the table above. You then check they are really different, tighten them to the same level of detail, and write the recommendation.
Idea, Device, Grammar (put these on every direction)
Add three rows to each direction: Idea (what the viewer should believe, 8 words or fewer), Device (one object carried through every shot, ending as the mark if it can) and Grammar (a rule the form obeys, for example it opens and closes on the same tick). Fill them from a list of real specifics (facts, real UI, real numbers), not from a look. Kill tests before a direction is written up: ownership, deletion (remove the device: does it still work?), specificity, "could a competitor's film say this?". Directions that differ only in palette fail the existing difference test and these. Cheap tells to avoid: a borrowed prop for an idea, a demo with no cause, a recycled end card, the device dropping out at a seam. Why: stops style posing as idea. Evidence: doc-level with four worked concepts; strength of the resulting films only claimed.
Hook options beyond the felt moment
- A story with a breaking point: one quantity builds over N steps and fails visibly at step N (five glasses fill with risk, the last overflows). Single screen, loopable, input to visible transformation to output. Evidence: two builders found it the one reliably interesting pattern (claimed).
- Music-led films: beat map first (BPM, beats, beats before the drop, ranges); the drop is the anchor.
- Real assets: a real capture or the real component beats redrawn UI.
Pacing reference numbers and the full rules: motion-studio/references/direction.md.