Pre-production is where videos are actually won: the cheapest edit is on paper, and the cheapest reshoot is the one the shot list prevented. Each letter is a gate; most chaos traces to a missing C (words-only scripts) or N (no shot list, so no batching and no clean edit).
S — Start from the spine
Before any words: the idea's one job (from idea-generation-and-ideation), the audience, the format
target, and a beat outline — hook / turns / payoff / loop for short-form (the retention structure itself
belongs to short-form-video-script / the long-form skills; this skill turns it into production). A script
without a spine produces footage without a point.
C — Columns: audio + visual
Write the two-column AV script — the broadcast standard. Left column: everything heard ([VO], dialogue,
[SFX], music cues). Right column: everything seen, beat by beat — B-roll, punch-ins, text pops, props,
angle changes. The rule: if the visual isn't written, it will be improvised, and improvisation defaults to a
static face. The visual-change-every-~2–4s rhythm is planned here and executed later by capcut's EDITS.
Tags ([VO]/[SFX]/[Action]) keep it machine- and human-readable.
E — Envision every beat
The storyboard is a decision document: what is on screen at every beat? Pick the cheapest form that
answers it — a shot table (fastest), rough boxes/stick figures, or AI previz frames when a client or
team needs to see it (Boords/Storyboarder.ai-class, or flux frames). AI-frame craft: explicit camera
intent per panel (the model guesses otherwise), locked palette/tone, a 3–5-line character bible + one
reference image for consistency, micro-beats for complex scenes — and boards stay directional, never
mistaken for final art. Skip boards honestly where they add nothing (one talking-head setup needs a shot list,
not twelve panels). For AI-video sequences, each panel is the generation's shot brief (→ luma) — and
previz frames cost cents where generations cost credits: board first.
N — Number the shots
Convert the board into a numbered shot list: shot #, setup (location/framing/lens/light), subject/action, audio pairing, props, est. duration. Then the leverage move — regroup by setup, not by video: that's what turns two weeks of content into one afternoon (sequence setups to minimize changes, batch outfit swaps, buffer takes on hooks, keep each video's own hook/CTA marked so the batch doesn't blur). Run the runtime math (~130–150 wpm; ~75 words ≈ 30s; pauses cost seconds) and a feasibility pass — can this actually be shot with these resources?
E — Edit on paper first
The final pre-production gate: a table read with a stopwatch (cut to the target runtime by collapsing the
middle, never the spine), a logic pass (does every visual earn its beat?), and the honesty pass (no staged-
as-candid moments, no lifted scripts — structure study yes, plagiarism no). Then hand off: the labeled shot
list to the shoot day, the footage map to the edit (capcut/descript), and the finished video →
scheduling-and-queue → WoopSocial. The human shoots, judges, and approves — the plan is the agent's;
the camera is not.
