Scene Design Prompt Skill

1. Purpose

Convert a user-confirmed scene specification into a precise, platform-independent English image-generation prompt for a single CINEMATIC PRODUCTION DESIGN BOARD.

The skill focuses on:

• spatial layout and scale
• spatial system design guidance for all environments (offline reference: enclosure, zoning, circulation, depth, extension)
• functional logic guidance for all environments (offline reference: primary / secondary / support functions, zones, adjacency, workflow, orientation, storage, privacy, visibility)
• identity logic guidance when the scene has an owner / user (offline reference: status, authority, wealth, taste, access, furnishing density, personal objects, behavioral traces)
• era / period and cultural context
• primary visual anchor selection
• spatial relationships between major elements
• 16:9 fixed nine-panel board composition
• spatial continuity across all panels
• lighting and atmosphere consistency
• period visual language guidance for Chinese historical environments (offline reference, no auto web research)
• reference-board framing and readability

The skill produces prompts and concise supporting text only. It does not generate images or videos.

Resources

Workflow

Follow this interaction flow in order. Steps 1 and 3 use plain-text questions; Step 2 uses the AskUserQuestion tool. The user is a Chinese speaker, so all guiding questions and option labels are written in Chinese.

Step 1 — Gather scene information

Ask the user a single guiding question to collect the essential information in one turn:

> 请描述你想设计的场景：是什么空间（如书房、寺院大殿、老式理发店）？什么时代和文化背景？有没有一个想作为主视觉参照物的物件或空间元素？它和其他元素大致的位置关系是怎样的？

Invite the user to answer all fields together; do not ask field by field. The primary visual anchor is optional in this step. If the user omits a field that would materially change the design (such as era or spatial type), follow up on that specific field only.

From the user's answer, extract a five-field Scene Brief (do not generate furniture yet — the brief only pins down the reasoning inputs):

This brief is a fixed reasoning input for Steps 2 and 4. It does not add a new interaction step — Step 1 remains a single question.

Step 2 — Confirm the primary visual anchor (conditional, use the AskUserQuestion tool)

Case A — the user already specified an anchor. Adopt it directly. Do not run the tool. If the user did not state the anchor's spatial relationship to the other major elements, ask one follow-up question:

> 这个锚点大致位于空间的什么位置？和门窗、主要家具的关系是怎样的？

Case B — the user did not specify an anchor. Derive candidates before running the tool:

1. Read references/anchor-selection.md and evaluate the described space against the LEVEL 4 / 3 / 2 hierarchy. Prefer elements that naturally organize the space's function (see FUNCTIONAL ANCHORING in references/functional-space.md): strong spatial presence, clear relationship to surrounding zones, stable location, repeated visibility — not small decorative objects.
2. Select the THREE strongest candidates according to VISUAL DISTINCTIVENESS × SPATIAL RELEVANCE × MULTI-VIEW RECOGNIZABILITY.
3. Present them with the AskUserQuestion tool. Each option label is the candidate element; each description states its hierarchy level and a one-line reason.

The user may pick a candidate, type a replacement anchor, or add spatial-relationship notes in the free-text input. Whatever the user confirms becomes the scene's fixed PRIMARY VISUAL ANCHOR.

Once confirmed, the anchor is the fixed spatial basis for the whole board: it must remain recognizable and spatially consistent across every panel.

Anchor decision outranks Production Design. The PRIMARY VISUAL ANCHOR confirmed here is a user-level decision. It has higher priority than any Production Design inference made later in Step 4: architecture, furniture, and soft dressing must all organize around the fixed anchor, and later design choices must never alter, replace, or demote the anchor to fit a preferred layout.

Step 3 — Confirm visual style

Ask one open question in plain text (do NOT use AskUserQuestion). The examples are suggestions only, not a closed list:

> 你想要什么呈现风格？比如超真实、日式动漫2D、皮克斯3D、美式卡通、Q版，也可以说别的。

Preserve whatever style the user specifies as authoritative. For the "超真实" style only, use the hyperrealistic annotation specified in references/board-spec.md; other styles use their own annotation, never the hyperrealistic one.

Step 4 — Build and output the prompt

Read the single template from references/templates/scene-design-board.md, fill every placeholder with the user's confirmed specifications, the confirmed anchor, and the chosen style. Keep the confirmed anchor fixed while all Production Design work (architecture, furniture, soft dressing) organizes around it — the anchor governs the design, never the reverse. Derive the SOFT DRESSING field (screens, hangings, textiles, fixtures, vessels, scrolls, display objects, small personal items, wall / column / window treatments) from space + function + identity + culture/period, not from a decorative checklist or a search-result list. Consult references/board-spec.md for the technical rules enforced by the template. Apply references/spatial-design.md to every scene when defining the environment's enclosure, scale, zoning, circulation, depth, extension beyond the frame, and camera-view descriptions (spatial logic applies to all cultures and periods). Apply references/functional-space.md to every scene when deriving the environment from its use: primary / secondary / support functions, functional zones and adjacency, circulation and access hierarchy, workflow, orientation, focal functional element, storage vs display, privacy and visibility — the space's form, objects, and circulation should emerge from what happens there (functional logic applies to all functions and cultures). When the scene belongs to or is used by a specific person (owner / user identity, e.g., a prince's study), apply references/user-identity.md to translate that identity into visible environmental consequences (spatial control, privacy, access, furnishing density, material quality, personal objects, behavioral traces) — the environment should appear to belong to the person, and identity must remain legible even with no people shown. For Chinese historical environments, additionally consult references/chinese-historical-environment.md for period-appropriate visual language (period groups, furniture height, social status, dressing, atmosphere, density, anachronism filter).

NO AUTO WEB RESEARCH. Do not search the web to verify scene elements by default. Rely on the offline references and general knowledge. Only perform web research when the user EXPLICITLY requests it (e.g., 用户主动要求查证/调研某个元素).

Then deliver the prompt in two ways:

1. HTML copy page — Read assets/prompt-copy-template.html, replace the {{PROMPT}} placeholder with the completed prompt (escape HTML special characters: & → &, < → <, > → >), write the result to an HTML file, and present it with present_files so the user gets a one-click copy button.
2. Chat text — Also show the completed prompt in the chat reply for direct review.

2. Core Constraints

The following constraints are enforced by the single template and must never be relaxed:

• Fixed board structure: EXACT 16:9 board, EXACTLY NINE rectangular panels, FULL WIDTH = LEFT 3/6 + CENTER 2/6 + RIGHT 1/6. See references/board-spec.md for panel definitions.
• One single environment: all nine panels describe the SAME coherent physical space. Camera changes the viewpoint, never the physical environment.
• Unique anchor: EXACTLY ONE PRIMARY VISUAL ANCHOR per scene, recognizable and spatially consistent across all views.
• Anchor outranks Production Design: the Step 2 anchor decision is a user-level input that overrides later design inferences; Production Design must organize around it and never alter or replace it.
• Soft Dressing is derived, not decorated: screens, hangings, textiles, lighting fixtures, vessels, scrolls, display objects, and small personal items are derived from space + function + identity + culture/period — never a decorative checklist or a search-result list.
• Continuity mandatory: consistent architecture, scale, furniture, doors / windows, materials, color, and lighting across all panels. Any conflict means generation failure.
• Natural daylight unless the script explicitly requires otherwise (e.g., a scripted night or candlelight scene).
• NO PEOPLE in any panel. Empty chairs or half-open doors implying recent departure are allowed; living characters are not.
• Spatial logic per references/spatial-design.md: the environment is designed as a spatial system (enclosure, scale, zoning, circulation, thresholds, depth, extension beyond the frame), not as walls + furniture; FRAME ≠ PHYSICAL BOUNDARY, and large environments must extend believably beyond the frame rather than being compressed to show all boundaries.
• Functional logic per references/functional-space.md: the space's form, zones, circulation, object placement, and traces must emerge from its primary / secondary / support functions; a functional label is not a visual style; establish function BEFORE decoration (function vs decoration ordering); anchor selection prefers functionally organizing elements.
• Identity logic per references/user-identity.md: when the scene has an owner / user, identity is expressed through a coherent system of consequences (spatial control, privacy, access, furnishing density, material quality, personal objects, wear and traces), never through stereotypes or luxury shorthand; distinguish status / authority / wealth / taste; identity stays legible with no people shown.
• Chinese historical scenes follow references/chinese-historical-environment.md: period-appropriate furniture and seating height, social status expression, dressing density, atmosphere cues, and anachronism filtering; prefer a conservative visual solution over inventing uncertain historical detail. No auto web research — search only when the user explicitly asks.
• No unnecessary Chinese text: include text in the scene only when the script explicitly describes it.

3. Information Priority

Resolve information in this order:

1. User-confirmed specifications
2. Existing project / scene specifications
3. Spatial logic from references/spatial-design.md (all scenes: enclosure, scale, zoning, circulation, depth, extension, camera & spatial truth)
4. Functional logic from references/functional-space.md (all scenes: primary / secondary / support functions, zones, adjacency, workflow, orientation, storage, privacy, visibility)
5. Identity logic from references/user-identity.md (when the scene has an owner / user: status, authority, wealth, taste, access, occupation consequences)
6. Period visual guidance from references/chinese-historical-environment.md (for Chinese historical scenes)
7. General design conventions
8. Conservative inference

Higher-priority information always overrides lower-priority assumptions.

If missing information materially affects the design, ask for clarification rather than inventing a major characteristic.

4. Quality Check

Before returning the prompt, verify:

• Correct template used (the single scene-design-board template)
• EXACT 16:9 board with EXACTLY nine panels (LEFT 3/6 + CENTER 2/6 + RIGHT 1/6)
• Spatial Master is TRUE 90-degree vertical overhead (ZERO TILT / ZERO PITCH / ZERO ROLL)
• EXACTLY ONE PRIMARY VISUAL ANCHOR, filled and unambiguous
• Three narrative coverage shots are clearly different viewpoints of the same environment
• EXACTLY FOUR UNIQUE assets, not duplicates
• Continuity constraints are preserved (architecture / scale / furniture / doors-windows / materials / color / lighting)
• Visual style matches the user's choice, without foreign style annotations
• No people in any panel
• Chinese historical scenes pass the period / status / dressing / anachronism checks in references/chinese-historical-environment.md
• Spatial design passes the FINAL SPATIAL TEST in references/spatial-design.md (enclosure, scale, depth, extension, sequence, zoning, hierarchy, circulation, layering, void, rhythm, continuity, spatial truth)
• Functional design passes the FINAL FUNCTION TEST in references/functional-space.md (purpose, zoning, adjacency, circulation, access, workflow, scale, orientation, storage, privacy, visibility, negative space, depth, continuity, realism)
• When the scene has an owner / user identity, the design passes the FINAL IDENTITY TEST in references/user-identity.md (role, status, authority, wealth, function, lifestyle, privacy, access, taste, visible consequences; no stereotyping)
• No auto web research was performed (unless the user explicitly requested it)
• No unnecessary Chinese text beyond what the script describes
• Prompt is platform-independent
• Unnecessary repetition has been removed

5. Output

Return the completed English prompt as the primary result.

Supporting explanation should be concise.

Do not reproduce the entire skill or reference material in the response.

The goal is not maximum prompt length.

The goal is:

spatial coherence + anchor consistency + continuity + visual accuracy