VIDEO

複数ショットのキャラクター連続性

人物の顔・体格・衣装・感情の流れを固定し、各ショットの動作、画面方向、カメラ接続を具体的に設計します。

PROMPT
Design a character-led cinematic sequence. Treat the character bible and continuity anchors as immutable visual truth. Each shot must have one clear performance beat and one motivated camera idea.

CHARACTER BIBLE
{{subject}}
Specify the adult character’s facial structure, apparent age, body proportions, hairstyle, wardrobe layers, footwear, accessories, carried props, dominant hand, baseline emotion, and distinctive marks.

LOCATION AND SPATIAL MAP
{{scene}}
Define stable landmarks, entrances, light direction, time of day, weather, foreground/midground/background, and the character’s starting position.

PERFORMANCE ARC
{{subject_action}}
Describe the character’s objective, gaze, expression, posture, gesture, locomotion, interaction, and emotional change. Every action must continue logically from the previous shot.

SHOT PLAN
{{shot_plan}}
For each shot state: framing and angle; character position and screen direction; one action beat; foreground/background activity; entry frame; exit frame; and the visual match or motivated cut connecting to the next shot. If the selected model is strongest at single shots, generate each numbered shot separately using the same references.

CAMERA LANGUAGE
{{camera_motion}}
Define a consistent lens family, camera height, support style, movement speed, focus transitions, and when the camera becomes still. Do not use a camera move merely for decoration.

LOCKED CONTINUITY
{{continuity_anchor}}
Repeat these anchors in every shot. Preserve face, proportions, costume construction, prop hand, color grade, key-light direction, geography, eyelines, and the 180-degree screen-direction rule unless a visible crossing shot intentionally resets it. No identity drift, wardrobe mutation, prop disappearance, axis jump, scale shift, or unexplained location change.

Negative prompt

identity drift, face mutation, age change, wardrobe change, warped anatomy, rubbery limbs, sliding feet, floating contact, object morphing, duplicate subjects, disappearing props, geometry drift, flicker, exposure pumping, texture crawl, background warping, teleportation, abrupt zoom, unmotivated cut, camera jitter, reversed screen direction, text, subtitles, logo, watermark

制作手順

  1. 被写体とシーンを先に定義し、次にカメラ、照明、スタイルを入力します。
  2. 最初の結果では人物同一性、形状、構図、主動作を確認します。
  3. 反復ごとに動作またはカメラ指示を一つだけ変更します。

動画プロンプトの原則

  • 動画の長さは生成設定で選び、プロンプト変数には入力しません。
  • 人物がいる場合は外見、表情、視線、姿勢、重心移動、手の動き、終了状態を記述します。
  • 人物がいない場合は前景・中景・背景の動きとカメラ経路を中心にします。
  • Image-to-Videoでは画像が外見を決めるため、テキストは動きを優先します。
  • 1クリップにつき主動作一つ、動機のあるカメラ移動一つから始めます。

検証

顔、手、製品形状、接地、画面方向、光、背景位置、ちらつき、フレーム間変形を確認します。