Scene Design

Scene Design Runtime

多参宗白梦客出品。禁止任何盗卖行为。

This is the lean runtime entry for scene design. The full manual remains available as on-demand knowledge, but the default path should produce a usable scene packet fast.

Core principle: scene is not background. Scene is the physical container of worldview, emotion, blocking, and story consequence.

For multi-shot or multi-segment AI video, distinguish the base location from the current scene state. The base scene preserves world identity and fixed anchors; the scene state preserves the specific shot's door/window status, traces, actor zone, camera opportunity, action lane, and before/after continuity.

For live-action overseas short dramas, scene design must become a production system: power-space architecture, 9:16 vertical blocking, recurring scene state, and video readiness. Read references/live-action-shortdrama-scene-system.md when the request mentions 欧美真人短剧, overseas short drama, scene library, Midjourney scene refs, vertical drama locations, boardroom / hospital / school / HOA / gala scenes, or asks whether scenes need experiments.

Output Contract

For most tasks, produce a compact scene packet:

When scene continuity is important, also include a scene continuity ledger: base scene identity, fixed spatial map, door/window state, wall/surface state, ground traces, light/weather state, actor zone, camera axis, protected anchors, allowed change, and next scene state.

If the user asks for a simple prompt, include a prompt-ready scene anchor block after the packet. If they need a reusable prompt packet, scene reference suite, or model-specific image prompt, hand off to /Users/baimengke/.agents/skills/prompt-framework/references/scene-image-prompt.md.

For Seedance / 即梦 multi-reference video, do not treat one physical place as one unchanging @Scene across every 15-second segment. Return a base scene plus per-segment scene states when the camera, actor zone, door state, wall marks, ground traces, light, or continuity residue changes.

For 胡金铨 / King Hu / 70年代武侠 / 文人武侠 scene mood, keep the scene packet mandatory and call /Users/baimengke/.agents/skills/prompt-framework/references/midjourney-wuxia-aesthetic.md only as a style/composition layer. The old knowledge base already has foreground obstruction, negative space, doorway light, side-backlighting, smoke/light rays, and earth/grey-blue palettes; the wuxia module simply organizes those atoms into a reusable cinema pack.

For multi-director scene fusion, use /Users/baimengke/.agents/skills/animation-studio/references/director-style-fusion-rules.md first. This skill only turns the chosen style lanes into space, light, color, material, camera opportunities, and continuity anchors.

Plot Upstream Bridge

Use the installed story skills as upstream diagnosis when the user provides raw story text, an IP outline, a sequence of beats, or asks for scenes that carry plot instead of only atmosphere.

Translate plot findings into physical space:

• A plot point becomes one scene task: establish, reveal, pressure, turn, release, or callback.
• Conflict becomes distance, height, threshold, blocked path, hard light, crowd pressure, or object ownership.
• Relationship power becomes who controls the entrance, center, seat, doorway, window, or highest platform.
• A hook becomes the first readable image: confrontation, forbidden object, abnormal trace, open door, broken rule.
• A cliffhanger becomes a changed scene state the next segment must inherit.
• World rules become fixed anchors: architecture, signage, ritual objects, work surfaces, weather, light source, or material wear.
• Story facts become scene anchors only when they affect action, blocking, traces, light, props, or before/after continuity.

Do not design a beautiful empty setting. Every chosen anchor should help a character act, fail, hide, collide, discover, or change. Do not import short-drama assumptions unless the user explicitly asks for short drama; default to animation, film, concept, ad, or asset-production logic.

Runtime Flow

1. Define the narrative task.

   • Establish world, reveal character, establish relationship, create tension, release emotion, mark a turn, or close a theme.

2. Translate task into space.

   • What layout, height, distance, entry, exit, obstacle, or empty area makes the task physical?
   • A scene must affect action, not only mood.
   • For short drama, identify who controls the door, desk, seat, bed, podium, document, or light.

3. Choose actor zones.

   • Where does the character enter?
   • Where can they act?
   • Where do they fail, hide, collide, or land?
   • What stays fixed so movement reads?
   • For 9:16, lock foreground pressure object, center actor lane, background power anchor, threshold axis, and handoff zone.

4. Add narrative objects.

   • Use 1-3 props, marks, or traces that reveal world, habit, danger, relationship, or change.
   • A prop should be able to cause or clarify an action.

5. Lock light, color, and material.

   • Light directs attention and emotion.
   • Color sets first emotional read.
   • Material tells the world rule through texture and use.

6. Handoff to storyboard or AI video.

   • Reduce the scene to stable anchors.
   • Name what must not change.
   • For multi-segment video, name what has changed in this segment's scene state.
   • Use @scene-name consistently across storyboard and prompts.

7. For overseas short-drama production, run the scene system gate.

   • Build a genre scene pack before writing episode prompts; keep recurring locations small and reusable.
   • Assign each location a power mechanism: corporate, legal, medical, education, civic, church, gala, or family wealth.
   • Create a scene state ladder: normal order, pressure setup, humiliation, reveal, aftermath, power reversal.
   • Test core locations with a 5-10 second motion audition before labeling them scene_video_ready.
   • For the detailed method, read references/live-action-shortdrama-scene-system.md.

Narrative Task Map

Scene Prompt Anchor Block

Use this when handing to image or video models:

[@SceneTag], [space type and layout], [fixed anchor], [actor activity zone], [narrative prop], [light source/direction], [color/material rule].
Keep consistent: [3-5 anchors].
Avoid: [forbidden drift].

For multi-shot scenes, repeat the @SceneTag, fixed anchor, and activity zone before changing camera language.

Space-To-Action Checks

• Does the scene create a consequence for the character?
• Is there a clear entry, exit, and activity zone?
• Is the most important action staged in the clearest light or silhouette?
• Can the model preserve the setting with 3-5 anchors?
• Does the next scene inherit visible evidence from the previous scene before changing anything?
• Does a prop or mark reveal story without dialogue?
• Does the scene say too much, or leave useful space?
• Does it connect to the previous and next scene?
• For vertical short drama, is there a clear 9:16 actor lane and readable first 3-second power image?
• Does the scene have a specific institutional or social power function, not just luxury decor?
• Can a character enter, be blocked, hand over a document, reveal evidence, or reverse power inside this layout?
• Are door, window, desk, bed, podium, or table positions stable enough for a video test?
• Does the scene state preserve traces and changed object positions instead of resetting clean?

Deep References

Read references/full-manual.md only when needed. Search these section names:

Animation-Studio Handoff

When used inside animation-studio, return the compact scene packet and prompt anchors. Let animation-studio decide story structure, character packet, shot timing, and model segmentation; let prompt-framework turn anchors into stable SCENE_ or SHOT_ prompt packets when needed.