Motion
Action lands better when you write timing, result, and final state, not just a list of verbs.
The model responds better to motion that has weight: inertia, recovery, almost hitting something. In a short clip, one clear physical goal is usually more stable than many tiny actions happening at once.
- Try to write who does what, when it happens, what result follows, and what state it ends in.
- In chase scenes, keep screen direction consistent and avoid flipping left and right without reason.
- Hands, faces, and logos break easily under hard motion. When the action is violent, simplify those details instead of leaving all the risk to the model.