ai-video-script-sop-remotion-diffusion

Structure the script as a shot execution table where each shot specifies duration, technique (diffusion or code), a five-part prompt (style anchor, micro-timing, entities, camera physics, transitions), and audio. Keep diffusion segments at 10 seconds or less with integer durations.

Use diffusion for photoreal scenes, complex lighting, and character close-ups requiring texture and mood. Use code (SVG/TSX/HTML/Remotion) for cartoons, vector motion, text, formulas, and charts where precise typography matters. Never mix both in a single shot.

Count the characters in each line and divide by the speech rate of about 4.5 characters per second, then round up to an integer. Verify the total of all segment durations equals the total audio duration before final rendering.

Use image-to-video generation where the previous shot's end frame becomes the next shot's start frame. For code-to-diffusion transitions, export the last code frame as the first diffusion frame, and maintain consistent camera movement direction across shots.

Diffusion clips are capped at 10 seconds per segment and struggle with on-screen text, charts, and precise camera control. They also cannot handle abstract metaphors directly, so prompts must convert concepts into concrete physical objects and explicit transitions.