Storyboard Decomposition Guide

Decomposition Principles

Shot Elements

1. Shot Title: Summarize the core content in 3-5 characters (e.g. "Waking up from a nightmare")
2. Time: Specific time + light description
3. Location: Complete scene description + spatial layout + environmental details
4. Shot Size: Long shot/full shot/medium shot/close shot/close-up
5. Angle: Eye level/low angle/high angle/side view/back view
6. Camera Movement: Fixed/push in/pull out/pan/track/dolly
7. Action: Who + specific operation + body details + expression
8. Dialogue: Complete dialogue for this shot
9. Frame Result: Immediate consequence of the action + visual details
10. Atmosphere: Light + color tone + sound + overall atmosphere
11. Duration: 10-15 seconds per shot
12. Static Frame Prompt:

    image_prompt

    , used for generating first frame/last frame/shot images
13. Video Prompt:

    video_prompt

    , video generation description segmented by 3 seconds (required)
14. Background Music Prompt:

    bgm_prompt

    , describes the suitable soundtrack style for this shot
15. Sound Effect Prompt:

    sound_effect

    , describes the key ambient sound/action sound of this shot
16. Scene Association: If it matches an existing scene,

    scene_id

    must be filled in
17. Character Association: Fill in

    character_ids

    to bind 0 or multiple characters involved in the current shot

Video Prompt Format

video_prompt

0-3秒：<location>咖啡厅</location>，近景，<role>小明</role>低头看手机，表情焦虑。
<n>3-6秒：<location>咖啡厅</location>，全景，门铃响，<role>小红</role>推门走入。
<n>6-9秒：<location>咖啡厅</location>，中景，<role>小红</role>微笑走向小明，坐下。

• <location>location</location>

  — Scene marker

• <role>character name</role>

  — Character marker

• <voice>character name</voice>

  — Voiceover/narration marker

• <n>

  — Time segment separator

Usage Steps

1. Call

   read_storyboard_context

   to read scripts, characters, scenes, and existing storyboard summaries
2. Complete shot decomposition based on the script first to ensure reasonable total duration and narrative continuity
3. Fill in complete fields for each shot:

   title / shot_type / angle / movement / location / time / character_ids / action / dialogue / description / result / atmosphere / image_prompt / video_prompt / bgm_prompt / sound_effect / duration / scene_id

4. Call

   save_storyboards

   to save the complete storyboards at one time
5. If adjustment is needed, call

   update_storyboard

   to modify specific shots

Scene Association Rules

• Prioritize using

  scenes

  returned by

  read_storyboard_context

• When

  location + time

  can be clearly matched, the correct

  scene_id

  must be filled back
• Do not generate non-existent scene IDs out of thin air
• If the script content clearly falls into an existing scene, do not repeatedly create new scene descriptions

Character Binding Rules

• character_ids

  must be selected from the character list returned by

  read_storyboard_context

• A shot can have no characters, or bind multiple characters
• As long as there are characters who clearly appear, are seen, perform actions or speak in the shot, they should be bound
• Pure environmental shots, empty shots, and object shots can pass an empty array

Quality Requirements

• description

  should be human-readable,

  video_prompt

  should be suitable for model generation, and the two should not replace each other

• image_prompt

  should highlight single-frame composition, character appearance, environment and light

• video_prompt

  should highlight time progression, action changes, and camera language

• bgm_prompt

  and

  sound_effect

  can use concise phrases, but should not be too vague such as only "tense" or "sad"
• If there is narration, write it into

  dialogue

  uniformly, in the format of

  Narration: content