| name | screen-to-video-bridge |
| description | 需要将剧本素材转化为受限的、可用于生成或预可视化的视频摘要时使用。 |
Screen To Video Bridge
Use this skill when the user wants to turn a scene, ad beat, or adaptation fragment into a screen_to_video_brief.
Typical cases:
- bridge a scene into a Sora / Seedance / generic video-generation brief;
- prepare a previz-facing visual brief without flattening the original screenplay;
- split a scene into clip-sized units with invariants and control intensity;
- preserve dramatic intent while adapting to short clip duration and execution limits.
Workflow
- Identify the source container: scene draft, scene card, ad beat, or adaptation fragment.
- Preserve source spans or evidence anchors whenever possible.
- Extract the dramatic function and minimum non-negotiable beats.
- Decide whether the work should stay one clip or be split into a clip chain.
- For long material, batch by structure first, not by arbitrary chunk size.
- Choose one main action and one main camera logic per clip.
- Return a layered brief with invariants, avoid rules, grounded visible assets, and runtime-specific notes only when they materially matter.
Output Contract
screen_to_video_brief: a bounded bridge document with source-scene intent, clip structure, shot/action logic, invariants, avoid items, and runtime-facing notes.
- The brief should not replace the screenplay.
- Tool details belong only when they change execution decisions.
Posture-Adaptive Guidance
When certainty = lost:
Ask: "What is the single most important visual moment in this material — the one image that cannot be lost in translation?" One anchor image is enough to build the brief. Do not ask for a full source summary.
When source = construct:
Primary activation context. Apply the full workflow: identify source container → preserve source spans → extract dramatic function → decide clip vs. clip chain → choose main action and camera logic per clip → return layered brief with invariants and avoid rules.
When certainty = certain:
Apply full output contract strictly. The brief must not replace the screenplay. Tool-specific details belong only when they change execution decisions materially.
When focus = world:
Visual grounding in world-specific assets (locations, objects, light qualities) is the primary concern. Invariants should include world-consistent visual markers, not just character appearance.
When focus = audience:
The downstream execution audience — model prompt consumer or previz team — is the real audience here. Calibrate specificity to what they need to make decisions, not to what is interesting about the source.
References
wp.screen-to-video-brief
rb.screen-to-video-brief
ka.cross-protocol-referral-edges
ka.long-script-batch-planning
ka.multilingual-visual-vocabulary
ka.prompt-delegation-levels
ka.screenplay-to-video-boundary
ka.source-span-traceability
ka.video-generation-shot-economy
ka.visible-asset-grounding