Principles
- Humans write the story, the model produces the visuals.
- 5–15 seconds per unit; the model has no cross-shot memory.
- Give less freedom: a corner > an entire room; an anchor > an empty set; one shot, one thing.
Structure
Logline： who · situation · desire · obstacle · cost
Short film: Hook → Setup → Turn → Climax → Resolve
Scene breakdown card:
场号 / 地点 / 时间 / 出场@
功能：建立|冲突|揭示|过渡
欲望→阻碍→结果
观众必读 1–3 事
Shot list：
镜号 | 秒 | 景别 | 唯一运镜 | @角色 | 动作动词句 | 对白 | 难度
Lock within the scene: key light direction, lens character,the side the camera never crosses the line to。
GEO (drafted in this module, pasted across the whole scene)
Write only space, not people; reuse verbatim for every shot in this scene:
GEO
— 地标与相对位置（中左/中右…）
— 180° 轴：相机始终在 ___ 侧
— 主光来自 ___
For positions use only frame-left / frame-right / N meters from the landmark, do not write "to the left of the hero".
Generation-oriented shot writing
| Technique | Method |
|---|---|
| Establish in the first second | For key scenes, spend the opening ~1s pinning only blocking/props/light; can pick up the tail of the previous shot's dialogue |
| One shot, one thing | Split door-break/entrance/landing; do not cram complex actions into the middle of "walk over then do" |
| Start already in state | "Already mid-swing / the door is already cracked"; no lengthy wind-up |
| Count in group scenes | Fix the number of people; a crowd can be a single asset, lock 1–2 leaders separately for close-ups |
| Two-space seam | Doorway + light-color contrast on both sides |
| Out of scale | Human body reference + visible proportional evidence |
| 10–15 times | Still bad → simplify the shot(split/reduce/swap angle), do not just add words |
Precise blocking, first-frame English lock sentence → validated by the storyboard module; final text → prompt module.
Director Brief → prompt
BEAT / CAST@ / LOC+GEO
FIRST FRAME：谁在哪、看哪
ACTION：按秒，每拍≤3 短句
CAMERA：唯一主运动
LIGHT：单源
PHYSICS / ACTING（欲望·阻碍·可见节拍）/ AUDIO / COUNT LOCKS
Acceptance
- Each scene desire→obstacle→result
- GEO written and shared across this scene
- Each shot has new information or emotional progression
- Difficult shots have an iteration budget
- Dialogue is performable, not expository