World Concept Art

Triggered when concept-art/ already exists and world-building has added new locations or changed world descriptions that affect visuals. Append new entries without rewriting existing ones, unless the visual foundation (color palette, lighting system) has changed — in that case, update overview.md and note which existing location prompts may need refresh.

Step 1 — Read Worldview Files

Read from worldview/[slug]/:

• world.md — extract from 世界概览 / World Overview and 世界规则 / World Rules:
  • Geography and terrain types
  • Climate and weather
  • Magic/tech visual manifestations (how does the world's power system look?)
  • Time period and architectural era
  • Overall mood and tonal cues
• locations.md — extract from each location entry:
  • Each named location's description
  • Architectural style and materials
  • Atmosphere and function
  • Any faction-specific visual markers (banners, symbols, lighting)
• factions.md (optional) — scan for faction-specific visual markers (banners, symbols, architectural influence) that affect location appearance

If concept-art/overview.md already exists (Mode 2), read it to maintain consistency with existing visual decisions.

Step 2 — Visual Element Extraction

Systematically extract these dimensions from the text:

For anything not described in the source text, leave it out — don't invent visual details that the world-building hasn't established.

Step 3 — Generate Prompts

overview.md — Visual Foundation

Write a concise visual specification that anchors all location prompts:

# [世界名称] — 视觉总纲 / Visual Foundation

## 色调基底 / Color Palette
[Primary colors + accent colors + mood associations. E.g., "青黛+云白+金霞; night scenes add 月蓝"]

## 光线系统 / Lighting System
[Default lighting + time-of-day variants + special effects. E.g., "Default: soft diffused light through cloud layers; Dawn/dusk: golden sidelight; Night: moonlight + floating spirit particles"]

## 氛围常量 / Atmospheric Constants
[What's always in the air — fog, particles, weather patterns. E.g., "Thin mountain mist always present; spirit energy manifests as faint floating light dots"]

## 风格锚定 / Style Anchor
[Art style keywords that apply to all prompts. E.g., "Chinese ink painting meets cinematic realism; ethereal, serene, vertical composition emphasis"]

## 可组合关键词 / Composable Keywords
- **背景层 / Background:** `[reusable background snippet]`
- **光线层 / Lighting:** `[reusable lighting snippet]`  
- **氛围层 / Atmosphere:** `[reusable atmosphere snippet]`

locations.md — Per-Location Prompts

For each location found in locations.md, generate multi-angle prompts:

### [地点名] — [English Name]

**全景 / Wide Establishing Shot:**
[prompt: far view showing the location's scale and environmental context]

**中景 / Medium Environmental Shot:**
[prompt: closer view showing architectural details and atmosphere]

**室内或近景 / Interior or Detail Shot:** (if applicable)
[prompt: interior environment or signature detail close-up]

**夜景变体 / Night Variant:** (if the location has a distinct night identity)
[prompt: same location under night lighting]

**情绪变体 / Mood Variants:**
- 平静 / Calm: `[prompt — serene, peaceful version]`
- 紧张 / Tense: `[prompt — stormy, ominous, red-tinted]`
- 史诗 / Epic: `[prompt — golden hour, dramatic scale, cinematic]`

**可组合片段 / Composable Fragments:**
- 背景层: `[background-only prompt, appendable to character prompts]`
- 光线层: `[lighting keywords for this location]`
- 氛围层: `[atmospheric particle/weather keywords]`

Prioritize locations that are:

1. Explicitly described in locations.md with enough visual detail to work from
2. Likely to appear in character scenes (check if any character scenes reference them)
3. Visually distinctive (skip generic locations like "a road" unless they're important)

atmosphere.md — Time/Weather/Event Prompts

Environment prompts not tied to specific locations — these are mood/condition overlays:

### [时间/天气/事件] — [English Name]

[prompt: atmospheric condition that can be layered onto any location]

**适用场景 / Use Cases:** [when to use this — e.g., "tension scenes", "peaceful moments", "world-level crisis"]

Cover at minimum:

• 2–3 time-of-day variants (dawn, midday, night)
• 1–2 weather conditions relevant to this world
• 1 special event atmosphere (if the world has notable visual events — festivals, magical phenomena, etc.)

details.md — Micro-Level Visual Elements

Close-up prompts for world-specific objects, textures, and symbols:

### [物件/纹饰名] — [English Name]

[prompt: detailed close-up of this world-specific visual element]

**用途 / Usage:** [where this appears — architecture, clothing, artifacts, etc.]

Only generate entries for items explicitly mentioned in the world files. This file grows incrementally as the world develops.

Output Structure

worldview/[slug]/concept-art/
├── overview.md      ← visual foundation (color palette, lighting system, style anchor, composable keywords)
├── locations.md     ← per-location multi-angle prompts (wide, medium, detail, night, mood variants)
├── atmosphere.md    ← time/weather/event atmospheric overlays
└── details.md       ← world-specific objects, textures, symbols

Prompt Writing Rules

Language: All prompts in English — even for Chinese-language worlds. AI image models trained primarily on English captions.

One prompt = one image. Don't describe multiple views or time transitions in a single prompt.

Structure each prompt as: [composition/shot type], [main subject], [key details], [lighting], [atmosphere], [style keywords], [quality tags]

Always include from overview.md: the style anchor keywords and at least one composable keyword layer (background, lighting, or atmosphere) to maintain visual consistency across all generated images.

Platform adaptation: By default, generate prompts for Midjourney V8 Alpha (--v 8). Add --raw only when maximum prompt control is needed (technical/architectural shots, specific art style matching) — omit it for mood/atmosphere shots where MJ's aesthetic polish helps. If the user requests a different platform or the worldview is anime/二次元 style, adapt prompts per references/platforms.md. Supported platforms: MJ V8, Niji 7, Seedance (video), Nano-banana Pro, SDXL. See references/platforms.md for full syntax, parameters, aspect ratios, and per-platform examples.

Don't over-specify. V8 handles detailed prompts better than previous versions, but still aim for 40-100 tokens per prompt. Leave room for the model's interpretation on non-essential details.

Incremental Update Rules (Mode 2)

When called after world-building expands an existing world:

1. New location added to locations.md → append its section to concept-art locations.md
2. Existing location description changed → update that location's prompts in place, flag with <!-- 🔄 updated: [reason] -->
3. World overview/rules changed with visual impact → update overview.md's affected sections (color palette, lighting, etc.) and add a note: <!-- ⚠️ visual foundation changed — existing location prompts may need refresh -->
4. New atmospheric phenomenon or world event → append to atmosphere.md
5. New world-specific object/symbol → append to details.md
6. No visual changes (only character/timeline additions) → do nothing, report "no visual changes detected"

After Writing

Report:

• Files created or updated
• Number of locations with prompts
• Any locations skipped (insufficient visual description) — suggest enriching them in locations.md
• Whether overview.md visual foundation changed (affects existing prompts)