Video Grammar
A locked structure for AI video prompts. This grammar treats the model like a crew: it needs a call sheet, not a poem. The blocks are always delivered in the same order, every prompt, no exceptions.
The 30-second architecture
30 seconds = two 15-second segments, six shots each, 0 to 2 seconds per shot. The first two seconds are the hook: one strong visual anchor that earns the rest of the runtime.
The eight mandatory elements
Every long-form video prompt must contain all eight. Missing one is the difference between a film and a screensaver.
- Hook design: the opening two-second visual anchor, named explicitly.
- Sound design: specific texture (fabric brushing stone, distant traffic hum), never "ambient sounds".
- Music progression: rise, shift, resolve. The music has an arc, not a loop.
- Dialogue or lyric cue: a spoken line, or
LYRIC CUE: [the emotion the line must hit]when lyrics carry the story. - Rule of thirds: a compositional note per key shot.
- Acting theory: Stanislavski emotional truth (what the character wants) or Chekhov physical gesture (one telling movement).
- Lighting mastery: motivated sources and a stated contrast ratio.
- Texture and motion: 8K-in-720p detail, smooth gimbal or drone motion, cinematic live-action realism only.
The block order (locked, every prompt)
Scene & Mood: [one or two sentences: what this moment IS, dramatically] Frame Map: [where each subject sits: left/centre/right third, foreground/midground/background, what negative space remains. Multi-shot: framing per shot.] Subject Lock (per character): [identity anchor + body orientation + pose + state + gaze + contact points + lock-down line. Trust the reference image for wardrobe; describe only what it cannot carry.] Cross-Frame Rules: [multi-character: never swap positions, never cross centre, never change depth; distance and screen sides held. Multi-shot: what carries across the cut.] Movement: [character motion + micro-motion + environmental motion, flowing paragraph, per-beat timestamps inline.] Last Frame: [the exact closing composition + on-screen text suppression line.] World Plate: [environment, era, weather, set dressing that matters.] Sound Bed: [diegetic audio only in this grammar: specific, physical, layered.] Capture Realism: [film-emulated, analog warmth in highlights, controlled blacks, real fabric, real skin, real haze, real grain.] Camera Capture: [one trimmed line: body/register, lens, movement style. Never doubled.]
Why this order works
Composition before motion: the model must know where everyone is before anyone moves. Subject Lock before Movement: identity is anchored before it is stressed. Last Frame stated explicitly: AI video drifts toward its ending, so define the ending. Camera as a single closing line: the capture spec is a stamp, not a paragraph; doubling it confuses the model.
@image1 … @image9), and attach them in exactly that order in the interface. Bullet 1 is always @image1. Break this mapping and characters swap wardrobe mid-shot.