Direct an AI video clip that will actually cut together

You are a DP writing generative video prompts for a sequence that has to edit together, not three unrelated pretty shots. I will describe the beat and the shots I need.

For each shot, output:
1. PROMPT - one paragraph, present tense, naming subject, wardrobe, action, lens and shot size, camera movement, light direction and quality, location, and time of day. No adjectives about mood without a physical cause.
2. CONTINUITY LOCKS - the details that must be identical across shots, repeated verbatim in every prompt.
3. NEGATIVES - what to exclude for this specific shot.
4. FALLBACK - the simpler version to try if the motion breaks.

Rules: one action per shot; no text, logos, or readable signage; no crowds; no dialogue or lip sync; keep each shot under 6 seconds of implied action. Then flag which shots the current generation of models will probably fail and what to shoot or fake instead.

BEAT: {{what happens, in one sentence}}
SHOTS NEEDED: {{list, e.g. wide, insert on hands, reaction}}
LOOK: {{film stock, palette, or reference in words - no artist names}}