Cinematic Shot Prompting

Overview

Use this skill when AI video generators produce warped backgrounds, perspective jitter, or morphing characters during dynamic camera moves. Video diffusion models (such as Runway Gen-3 Alpha, Kling AI, Luma Dream Machine, and Sora) do not have intrinsic 3D coordinate awareness; generic adjectives like "cinematic orbit" or "dramatic zoom" frequently lead to severe hallucinations.

This skill translates natural narrative descriptions into a formal 7-Slot Camera Motion Contract, providing explicit mechanical anchors, velocity paths, and parallax ratios that diffusion engines can reliably render.

Core Capabilities

• 7-Slot Motion Contract Compilation: Converts informal scene notes into physics-grounded camera vectors.
• Dolly vs. Zoom Decoupling: Explicitly separates physical lens position translation from optical focal length adjustments.
• Rigid Parallax Anchoring: Establishes occlusion velocity differentials between foreground occluders and background horizons to prevent background melting.
• Subject Bounding-Box Locking: Ensures consistent subject tracking without accidental character deformation.
• Terminal Frame Landing: Defines landing coordinates and terminal focal states.

The 7-Slot Motion Contract Specification

Every compiled shot adheres to this 7-point specification:

1. [Anchor Axis]: Fixed 3D origin coordinate in scene space.
2. [Optical Mechanism]: Physical translation (Dolly, Track, Crane, Jib, Orbital arm) vs optical zoom.
3. [Velocity Profile]: Linear continuous rate or ease-in / ease-out curve.
4. [Subject Lock]: Bounding-box tracking anchor (e.g. eye line, torso center) vs decoupled pan.
5. [Parallax Mechanics]: Relative motion velocity between foreground elements and distant horizon.
6. [Environmental Response]: Volumetric lighting shifts, lens reflections, and weather refraction.
7. [Terminal Landing]: Fixed end-frame composition, focal plane, and final framing state.

Workflows

1. Compile Camera Motion with CLI Tool

You can compile narrative intentions using the built-in compiler script:

# Natural language intent parsing:
python3 scripts/motion_contract_compiler.py \
  --intent "A lone detective walking in neon rain at night, camera orbits slowly around him" \
  --format markdown

# JSON output for automated agent pipelines:
python3 scripts/motion_contract_compiler.py \
  --intent "Rapid drone crane ascending over ancient mountain peaks at dusk" \
  --format json

# Built-in smoke fixture verification:
python3 scripts/motion_contract_compiler.py --sample --format json

2. Manual Agent Compilation

When invoked directly within Claude Code or Cursor, provide a shot idea:

"Direct this shot: A samurai standing on a mountain peak at dusk. The camera starts behind him and cranes upward into an epic high-angle wide view."

The skill formats the result:

[Camera Instruction]: Mechanical crane jib arm ascending vertically (+2.5m elevation) while tilting downward -30 degrees along central sagittal axis. Linear velocity 0.4m/s.
[Subject Anchor]: Samurai back profile anchored at bottom-center third. Silhouette remains locked and sharp against the setting sun.
[Parallax Mechanics]: High motion differential on flying cherry blossoms in immediate foreground; low differential on mountain ridge; zero parallax on distant cloud layer.
[Terminal Landing]: Settles at high-angle establishing wide shot, locking horizon line at top third.

References & Theoretical Foundations

Detailed cinematography principles, parallax velocity formulation, and diffusion engine mapping are documented in:

• Camera Mechanics Reference
• American Cinematographer Manual (10th Ed.), American Society of Cinematographers (ASC).
• Cinematography: Theory and Practice (3rd Ed.), Blain Brown.