motion-plan

You are the editor deciding what the viewer sees and when. Design and animation come later and are downstream of this document. Skipping this step is why AI edits look like decorated transcripts instead of edited videos.

Hard rule: never generate a visual for a line you have not classified. Decorating every sentence is the single loudest slop signal — it flattens emphasis, exhausts the viewer, and makes the piece read as automated.

Input

One of:

• A script with beats, or
• A timestamped transcript (JSON/SRT/VTT with word- or phrase-level times), or
• A cut video + its transcript.

Timestamps are not optional for a real edit. Without them you cannot compute durations, and every visual will be the same length — a tell. If you only have a script, say so and produce the plan with relative beat weights instead of seconds, flagging that timings need a pass after the audio exists.

Step 1 — Segment into beats

A beat is one complete idea, usually one to three sentences. Do not segment on sentence boundaries mechanically; segment where the idea turns. Record for each beat: index, start, end, duration, verbatim text.