skills/kling-ai-generate-video/references/scene-patterns.md

# Video scene patterns

Choose one primary pattern. Combine patterns only when the duration and live model can support a coherent result.

## Cinematic continuous shot

- Define opening composition, subject action, one camera path, environmental response, and ending beat.
- Anchor scale and spatial direction so movement reads clearly.
- Use for atmosphere, character moments, product reveals, transformations, and image animation.

## Image animation

- Treat the source as locked composition and identity rather than a broad mood reference.
- Describe credible motion: breathing, cloth or hair movement, gaze, hands, environmental effects, and restrained camera drift.
- Lock faces, body proportion, product shape, logos, background architecture, and color.
- Avoid large subject movement, aggressive camera movement, and major scene transformation in the same short clip.

## Product showcase

- **Studio reveal:** controlled orbit or push-in, designed reflection travel, clean surfaces, and a clear silhouette.
- **Function demo:** one function or action per beat, clear hand contact, believable mechanics, and no invented capability.
- **Lifestyle:** communicate context through action rather than unsupported copy.
- **Unboxing:** sealed state, opening action, reveal, reaction, and detail only when the user requests a multi-shot sequence.
- **Commercial ad:** hook, product evidence, and hero detail; leave slogans and end-card copy for deterministic post-production unless supplied.

## UGC and presenter-led content

- Use natural eye-level framing, credible phone movement, everyday settings, and imperfect but intentional performance.
- When the live model lacks requested synchronous audio controls, separate dialogue/narration planning from visual generation rather than claiming a completed voiced video.
- Never invent testimonials, personal experience, before/after effects, endorsements, or price.

## Explainer

- Give each visual beat one idea: introduce the problem, show the mechanism, demonstrate evidence, and close with user-supplied action guidance.
- Match visual complexity to available narration time. Prefer iconic scenes, diagrams, object metaphors, or controlled product demonstrations over unreadable text blocks.
- When final synthesis or audio is outside the live model's capability, deliver the generated clip or scene plan honestly.

## Multi-shot narrative

- Use multiple shots only when the user requests them or place, time, scale, or information state changes.
- Give each shot a purpose, duration, framing, action, camera path, and continuity anchors.
- Preserve clothing, faces, product structure, light direction, palette, screen direction, and location details.
- When explicit timing is required, shot durations must sum to the requested total.

## Social formats

- **9:16 vertical short:** reveal the hook immediately, keep action vertically centered, place faces/products in the upper two-thirds, and reserve lower UI-safe space.
- **1:1 square feed:** use medium framing, clear silhouettes, and restrained lateral motion. Choose any other ratio only from the selected model's live values.
- **16:9 campaign or YouTube:** use wider spatial orientation, stronger entrance or reveal, and intentional negative space.

## Concept selection

Offer three directions only when the user asks for brainstorming, alternatives, or concepts. When generation is already requested, select the best-fitting direction internally and continue; do not replace an authorized task with a proposal.

Possible directions are demonstration, emotion, and spectacle. For each brainstormed direction, explain the hook, shot structure, camera behavior, duration fit, and protected brand or reference facts.