地域文化造字海报

Regional Culture Poster

Turn a place into a cultural structure, not a landmark collage. Make one large Chinese character the composition's visual skeleton and connect it to one primary cultural structure, one supporting cue, and a restrained text system.

Current preferred baseline: the most successful direct images are the earlier regional-cities-10-direct-image2 style: one culturally condensed huge Chinese character, physically fused with a cropped architectural or spatial structure, with strong masking/negative-space tension and sparse copy. Minor direct-generation imperfections are acceptable when the image has strong concept, spatial force, and a memorable character-image relationship. Do not overcorrect into safe full-place-name wall posters.

Anti-rustic calibration: keep the single-character spatial collision, but make it contemporary. Avoid heritage-board, folk-tourism, non-material-culture-promo, ancient-rubbing, dusty-earth, and old-wall nostalgia unless the user explicitly asks for that mood. Prefer abstract construction, cool modern materials, controlled contrast, structural voids, glass/metal/concrete/water/light, and editorial distance.

Text-removal recognition calibration: the poster must not rely on the printed city name to carry the region. If the city label, theme line, and keywords were covered, the image should still contain at least one culturally specific and visually legible mechanism that points to the place: a distinctive building typology, craft process, climate behavior, street-life pattern, material system, water/terrain condition, ritual order, food/tool/workshop logic, or historical spatial structure. Generic modern materials such as glass, concrete, bridge, skyline, water reflection, and fog may support the design but cannot be the primary cultural evidence by themselves.

Two-mode calibration:

• MODE A / structural character poster: the Chinese character is the main visual body. City structures grow into the glyph through masking, negative space, material transition, and spatial cuts. Use this when the user wants 大字、造字、城市字体、字形实验、一个主字、直接图像生成, or the current single-character series direction.
• MODE B / regional editorial poster: the city image is the main subject, and the city name is an editorial anchor rather than the whole image container. Use this when the user wants stronger editorial feeling, richer text hierarchy, cultural research poster, magazine-like city poster, clearer cultural explanation, or says the big-character masking approach is too forced, too empty, too illegible, or too dependent on one huge glyph. Read mode-b-editorial.md before planning or rendering MODE B.

Route the request

Determine the requested stopping point:

1. Strategy only: deliver candidate research, direction selection, and the analysis card.
2. Prompt package: deliver strategy plus production-ready image prompts and typography directions.
3. Direct rendered poster: generate the complete flattened poster, including imagery, dominant character or editorial title, exact copy, masking, and layout, in one image-model call. Prefer this when a capable image model is available or the user asks for direct generation, one-shot output, no compositing, or no post-production.
4. Layered rendered poster: generate a visual foundation, add deterministic Chinese typography, reconnect image and type, then export. Use this when copy accuracy is contract-critical or direct generation repeatedly fails the text gate.
5. Batch series: plan all regions first, run global deduplication, then render the series in one locked render mode.

Do not render images when the user only asks for analysis, review, or prompts. When the user asks to create, make, design, or generate the poster, treat that as authorization to produce the requested artifact.

Establish inputs

Use supplied constraints. Otherwise apply these defaults:

• region: required; infer whether it is administrative, historical, or cultural
• count: 1
• aspect ratio: 3:4
• language: Chinese
• style: restrained, high-design, contemporary, quiet, editorial, exhibition-like; avoid commercial poster, movie-poster, tourism campaign, and decorative premium looks by default
• modern/traditional layering: allowed when culturally defensible
• supporting English: optional and subordinate
• sentence and 3-5 keywords: included unless the user asks for a text-free image
• render mode: prefer direct with a capable image model; use layered for copy-critical work or when explicitly requested
• visual mode: infer MODE A or MODE B from the user's taste feedback; default to MODE A for “城市字体 / 大字 / 造字”, and MODE B for “编辑感 / 杂志感 / 城市文化研究 / 图文丰富 / 不要只有大字”

For missing noncritical details, proceed with defaults. Ask only when the region itself is ambiguous enough to materially change the result.

Follow the workflow

1. Build the cultural candidate pool

Analyze at least these seven dimensions before composing:

• geography and climate
• architectural structures
• materials and local color
• historical and literary context
• everyday life and craft
• contemporary urban conditions
• spiritual temperament

Read cultural-analysis.md for the candidate vocabulary, selection rubric, and anti-stereotype rules.

2. Generate and select directions

Generate 3-5 internally coherent directions. Score them on:

• regional specificity
• cultural recognizability
• visual translatability
• potential to merge with character structure
• avoidance of cliche
• differentiation within the series
• text-removal recognizability: whether the place still has cultural evidence after covering the city name and explanatory copy

Select exactly one primary cultural structure, one supporting cue, and one temperament. Name the direction as 地区｜本次文化主题.

3. Choose and transform the main character

Prefer, in order:

1. one culturally condensed character, such as 京、粤、闽、蜀、秦、沪、苏、滇、藏
2. a short historical or cultural name, such as 北平、岭南、巴蜀、江南、三秦
3. the full place name only when shorter choices are misleading

Keep the character readable. Derive every deformation from evidence such as roof pitch, gable line, gate void, alley path, water curve, mountain contour, courtyard order, arcade rhythm, wall boundary, or old/new urban layering. Never decorate the character without a cultural source.

4. Define the visual system

Use:

• one dominant character as the visual body
• one primary cultural structure
• one supporting cultural cue
• generous real whitespace
• one dominant color, one or two supporting colors, and one neutral
• cropped structures, material interfaces, paths, edges, silhouettes, and light
• masking, negative space, interruption, layering, or partial occlusion between image and type
• exhibition-key-visual logic: abstract spatial section, material cut, wall aperture, threshold, compression, void, or civic infrastructure fragment may carry the idea more effectively than a recognizable building face
• one non-substitutable local mechanism: choose one thing that would be difficult to swap into another city without becoming false, such as Beijing palace-wall axis, Guangzhou arcade-river humidity, Nanjing plane-tree city-wall aperture, Shanghai lane-house stone-gate compression, Foshan ceramic kiln/workshop frame, Chongqing stair-bridge cliff section, Harbin ice-brick winter street, Suzhou garden-window water borrowing, or Quanzhou maritime temple/arcade layering
• prefer one culturally condensed dominant character when it can remain broadly readable and conceptually stronger than the full place name; use the full place name as the dominant typography only when the single-character route becomes misleading, generic, or unreadable
• if choosing the full place name, keep it independent, exact, and immediately readable, then put creativity into space, material, light, cropping, and image logic; do not let this fallback make the poster safe or ordinary
• controlled custom typography as the main enrichment layer: adjust stroke weight, serif shape, counters, rhythm, cuts, terminals, alignment, spacing, and material edge behavior according to the place; enrich the title before adding extra imagery
• a clean structure may still feel designed when the title has a deliberate type system, such as stone-cut Song style, slab-like civic gothic, porcelain-blue serif, wall-painted sans, weathered inscription edge, or narrow vertical title stack
• contemporary anti-rustic material logic: cold grey, fog white, deep green-black, glass reflection, brushed metal, wet concrete, polished stone, abstract water plane, shadow cuts, and restrained neon/sodium accents may replace ochre wall, ancient paper, dust, old brick, heavy rubbing, and faux-historical texture

Avoid:

• full landmark views and postcard compositions
• tourism advertising, slogans, or check-in imagery
• icon collages and unrelated traditional patterns
• movie-poster drama, commercial campaign polish, centered heroic building facades, cinematic spectacle, and over-produced atmosphere that makes the work feel generic
• heritage-tourism poster mood, non-material-culture promo texture, faux-ancient rubbing, dusty ochre nostalgia, old-wall-only backgrounds, heavy stone inscription aesthetics, and generic museum souvenir styling
• making the main title hard to read in pursuit of clever character-architecture fusion
• solving emptiness by adding decorative motifs, symbols, extra props, or busy background detail when typography could carry the richness
• overcorrecting readability into ordinary full-name title posters, plain wall compositions, or safe typographic layouts with no strong cultural mechanism
• making the culture so abstract that, after covering the text, the image becomes a generic modern poster for any city
• using generic urban ingredients such as glass curtain wall, bridge, water, fog, skyline, concrete, neon, or reflective floor as the main cultural idea without a place-specific structure attached
• making a poster feel “traditional” by default through earth yellow, sepia, cracked plaster, rubbing texture, seal-like composition, or antique typography
• red-and-gold national-trend styling by default
• calligraphy plus seal as a generic shortcut for “Eastern”
• dense explanatory copy, cartoon rendering, and fixed templates that only swap names

Read visual-system.md before writing the final art direction or rendering.

For MODE B, read mode-b-editorial.md and follow its city-image-first workflow. MODE B must not silently fall back to the giant-character mask system.

5. Output the planning layer

Always present a compact candidate pool and 3-5 direction options, then emit this completed card:

### 地区文化分析卡
**地区：**
**本次主题：**
**主大字：**
**主文化结构：**
**辅助文化元素：**
**精神气质：**
**色彩建议：**
**构图建议：**
**关键词：**
**一句短句：**
**大字变形逻辑：**
**应避免内容：**

Keep the sentence poetic and concrete, not promotional or didactic.

6. Produce prompts or artwork

For prompt packages, output:

1. final image-generation prompt
2. character-transformation logic
3. image/type relationship
4. render-mode plan: direct generation or deterministic post-typesetting
5. A/B/C variants when multiple directions are requested

Read production.md for prompt construction and render-mode selection. For direct final generation, also read direct-generation.md.

For rendered posters:

1. Lock direct or layered before rendering; do not mix modes silently.
2. In direct mode, ask the image model to produce one finished flattened poster with exact copy, dominant character, cultural structure, masking, and layout together. Do not reserve placeholders or add local overlays afterward.
3. In layered mode, generate a clean visual foundation, add exact typography deterministically, and rebuild masking and overlap.
4. If the user explicitly asks for direct generation, fix failed glyphs or copy by regenerating the whole image. Do not repair it through compositing unless the user authorizes a mode change.
5. Inspect every exported image at full size and thumbnail size.

Save generated assets according to the active host environment or the user's explicit output path. For public or shared use, do not assume a Windows drive, personal directory, or private workspace path. If no output location is provided, choose a local project-relative output directory such as:

• images: ./outputs/images
• drafts: ./outputs/drafts
• temporary files: ./outputs/temp
• final exports: ./outputs/exports

7. Validate before delivery

Reject and revise results that fail any of these gates:

• The place is recognizable through cultural structure rather than a landmark inventory.
• Text-removal gate: if the city name, theme, and keywords are mentally covered, at least one visible structure, material, climate, craft, street-life, terrain, or spatial mechanism still points to the region. If not, revise before delivery.
• The main character is correct, readable, dominant, and transformed for a stated reason.
• When the full place name is the main title, every character is correct, independent enough to read instantly, and not dissolved into architecture, shadow, or texture.
• Prefer a strong single-character direct result with a few tolerable text/glyph imperfections over a fully clean but visually ordinary full-name poster, unless the user explicitly demands exact copy above all.
• The image feels like an exhibition key visual, art-book cover, or cultural research poster, not a movie poster, tourism campaign, commercial KV, or souvenir advertisement.
• In direct mode, the city, theme, keywords, epigraph, and English metadata contain no invented, duplicated, or malformed copy.
• The design has one center, strong hierarchy, and genuine whitespace.
• Clean compositions still contain a considered typography design: the dominant title has intentional stroke rhythm, spacing, edge treatment, or material logic rather than default unmodified text.
• Color, material, and atmosphere come from the place rather than decoration.
• The work feels contemporary rather than rustic: if the concept uses historical material, balance it with modern cropping, cool palette, abstraction, or spatial tension so it does not become a heritage-tourism board.
• The sentence reads like an exhibition epigraph, not tourism copy.
• The work avoids cliche national-trend styling and generic city-poster composition.
• A series entry differs from prior entries in structure, character logic, palette, or composition.
• The delivered file preserves the locked render mode; a direct deliverable has no post-generated image or typography layers.

For batch work, read series-and-examples.md and perform a global deduplication pass before rendering.