dongfang-cover-design

Eastern Aesthetics Covers, Posters, and Square Images

Condense the theme into a single visual core, and use objects, people, products, text, color fields, light and shadow, and negative space to establish the viewing path. Eastern aesthetics is a method of composition, material, and imagery, not a subject-matter threshold. Poster mode is not limited to exhibitions; for commercial needs, visual direction determines information density, communication intensity, and the hierarchy of action information.

Inputs and Defaults

The only required input is the theme. Accepted: main title, subtitle, purpose, core object or person, cultural motif, aspect ratio, primary and secondary colors, visual intensity, specified focal-convergence method, reference images, formal information, atmospheric copy, layout, and run mode.

layout：auto | landscape-cover | portrait-poster | square-graphic, default auto。
run mode:create | prompt, default create. Run mode only determines whether to generate images and does not participate in landscape/portrait orientation judgment.
landscape-cover Default 16:9.
portrait-poster Default 9:16; also accepts 2:3, 3:4, 4:5, and approximate A-series paper ratios.
square-graphic Default 1:1, suitable for social media square images, celebration moments, brand communication, and product key visuals.
Visual intensity can be set to 克制高级 | 中等传播感 | high-density-thumbnail. When intents such as "high density, viral hit, thumbnail, feed cover" appear, automatically enable high-density thumbnail mode; it is a communication structure, not a fixed art style.
User-specified layout and ratio take priority; when the two conflict, the explicit ratio determines the orientation, and the plan should explain the normalization result.
When the user provides only the theme, autonomously complete the visual strategy without requiring a form.
prompt Deliver only the visual plan and final image-generation prompts;create use the same plan and invoke the built-in image generation capability.
All finished outputs by default and by requirement must use a fully opaque background; do not generate transparent backgrounds, Alpha backgrounds, or transparent canvases containing only a cut-out subject. The background should be part of the composition, color field, and space.
When the user provides reference images, first determine whether they are style references, composition references, or editing targets, and do not copy unrelated content without authorization.

Layout Routing

Determine the layout according to the following priority:

When the user explicitly specifies landscape-cover、portrait-poster or square-graphic , adopt it directly.
When the user explicitly gives 1:1, or intents such as "square," "square image," "block design," or "social media square image" appear, adopt square-graphic。
When the user explicitly gives a vertical ratio, or vertical intents such as "vertical," "vertical screen," "9:16," "poster," or "exhibition poster" appear, adopt portrait-poster。
For all other requests, adopt landscape-cover。

After the layout is determined, do visual direction. Choose appropriate visual types for content, culture, exhibition, and commercial subjects respectively, and do not mechanically apply museum templates. When high-density thumbnail intent appears, additionally read high-density thumbnail mode; for vertical, additionally read vertical poster system, and for square images, additionally read square communication image system; all three layouts share visual system。

Information and Copy Boundaries

First establish a "text source table," then design the text hierarchy.

User's original text: themes, titles, column names, brand names, product names, prices, discounts, stores, institution names, official exhibition names, event names, dates, locations, organizers, artifact names and periods, collection attribution, personal names, etc., preserved verbatim, with no speculation about missing content.
Verifiable facts: use only user input or trusted materials explicitly provided by the user. It is forbidden to generate fake prices, fake discounts, fake product specifications, fake brand promises, fake dates, fake addresses, fake institutions, fake archives, fake qualifications, fake returns, or fake contact information that would mislead people and potentially cause losses.
Automatic atmospheric copy: when the design requires it, add at most 1 Chinese atmospheric word or curatorial-style short phrase of 2–8 characters, and it must come directly from the thematic imagery. It may only carry emotion and narrative; it must not masquerade as an official name or factual statement.
Optional English atmospheric copy: add 1 short English line only when an international publishing feel is genuinely helpful, to carry the layout rhythm; it must not impersonate an official English name, institution translation, or event information.
Environmental atmospheric text: allow 1–3 groups of short text to appear on blackboards, signs, menus, packaging, flags, street signs, journals, or props as visual elements and world-building details. Readable or semi-readable is acceptable, but it must come from the thematic imagery and must not masquerade as the verifiable facts above.
Atmospheric copy may be reduced or removed for composition; the user's official text must not be rewritten for composition. Random classical poems, empty template phrases, and irrelevant English are still not generated; theme-related environmental short phrases do not count as "random small text."
In the visual plan, image-generation prompts, and final delivery, label "user-provided," "auto-added: atmospheric copy," and "auto-added: environmental atmospheric text." Fact fields unrelated to the theme need not be listed; only facts involved in the current context but not provided by the user are labeled "not generated."

Workflow

Determine layout and aspect ratio according to the layout routing.
Extract one cultural proposition from the theme and choose one absolute visual core. The core can be an object, a person, a landscape, text, or a metaphor that can be clearly visualized.
Read the visual guidance, and complete the type, shot scale, media fusion, three-layer space, dominant graphic, narrative action, text system, and complexity budget. The visual core does not mean defaulting to a large face or an isolated product close-up.
If high-density thumbnail mode is enabled, first establish "1 core + 1 movement + 3–5 auxiliary clusters + 2–4 colors + 1 title anchor" according to high-density thumbnail mode , then choose a clustered-core structure.
Choose one primary clustered-core structure: concentric structure, narrative reflow, Typography core compression, light-field focus, diagonal gravity, or framed convergence. Vertical formats can use vertical progression, top-bottom core compression, vertical framing, or a vertical light field. Add at most one auxiliary compositional force, and it must return to the same core.
Establish controlled richness. Normal mode retains 2–6 visual units with clear responsibilities; high-density mode can be split into 3–5 auxiliary clusters, but within each group the responsibilities are the same and the weights differ. Distinguish foreground, midground, and background; decide which elements must be removed to ensure it is still a single core at a glance in the thumbnail.
Let the title participate in composition and narrative. Text should enclose, compress the core, be coaxial, interweave, occlude, or become a negative-space boundary; it must not be attached as an independent UI label in blank space.
Establish a text source table, determine the hierarchy of formal information and atmospheric copy, then determine 2–4 primary colors, materials, light and dark, safe area, and viewing order.
Check against the shared visual system; for vertical formats, also check against the vertical poster system. When organizing final instructions and deliverables, read prompts and output contract。

Title strategy

Use the user's main title as-is. 4–8 Chinese characters are best suited as the visual main title.
When the title is long, you may distill a 4–8 character visual display title and use the full original text as the formal subtitle; you must list both and their sources, without changing the factual meaning.
When the user provides only a theme and no formal title, you may use the theme as a theme identifier, but must not claim on your own that it is an official exhibition name, event name, or brand slogan; automatically generated atmospheric copy must be labeled separately.
After generation, check the text character by character; if there are obvious wrong, missing, or extra characters, make only one targeted text correction while keeping the subject, composition, colors, and text-image relationship unchanged.
If after one correction the glyphs and composition still cannot both be satisfied, keep the version with better text-image integration and clearly state that manual text proofreading is needed.

`prompt` mode

Output strictly according to the reference file: layout and ratio, visual core, cultural metaphor, visual guidance card, focal concentration method, image path, text-graphic relationship, text sources, color and material, deletions, final image generation prompt.

The prompt should be independently executable, including canvas size, purpose, subject, composition, original text and sources, materials, lighting, colors, viewing order, and relevant prohibitions for this instance. Do not generate images, and do not output lengthy internal reasoning.

`create` mode

Organize production-grade prompts using the visual plan and call the built-in image generation tool; do not switch to an external CLI unless the user explicitly requests API/CLI.
After generation, check the theme, art direction, single-core hierarchy, focal concentration path, spatial layers, text sources and glyphs, materials, 2–4 colors, safe area, negative space, thumbnail recognizability, and confirm that the canvas background is completely opaque.
Iterate only on the most obvious problems; handle text errors according to the "Title strategy" and do not retry indefinitely.
Save landscape to output/dongfang-covers/; save vertical to output/dongfang-posters/; save square to output/dongfang-squares/. Use themed, safe file names; when names collide, add -v2、-v3, and do not overwrite original files.
Deliver the image, save path, layout and ratio, brief strategy, text sources, final prompt, and text proofreading status.

Acceptance criteria

Shared standards

At first glance, see the single core; at second glance, follow text, light, landscape, or color blocks back to the core; at third glance, read the theme information.
The subject silhouette and main title remain recognizable in the thumbnail, and key content does not touch the edges.
Genre, lens, media fusion, and dominant graphics form a unified art direction; it cannot be merely a high-quality photo with a title added.
Objects, figures, or landscapes are connected to the environmental light and shadow; the Eastern feel comes from composition, materials, and imagery relationships, not from piling up dragons and phoenixes, auspicious clouds, ink wash, and golden light.
All verifiable facts can be traced back to user input; all automatically generated text is clearly labeled as atmospheric copy.
The final product is a complete opaque canvas, and the background participates in the narrative and composition; there is no transparent background, alpha cutout, or isolated cutout subject.
The result has a professional level of finish consistent with its purpose: for cultural subjects, it can approach cultural magazines, feature key visuals, or art publications; for commercial subjects, it can be more direct and higher in contrast, but must not degrade into templated guochao, information piling, or cheap promotional visuals.

Additional standards for landscape orientation

Under the default 16:9, maintain a horizontal viewing path and platform safe areas; text, subject, and extended movement converge together, not mechanically centered.
Character-narrative covers retain at least two of environment, action, or symbolic props; a close-up of a face cannot replace narrative.

Additional standards for portrait orientation

The top, middle, and bottom layers have clear roles, and vertical viewing is natural; the bottom information area shows only real information provided by the user and relevant to the theme.
In a 9:16 thumbnail, the subject, title, and core color blocks remain recognizable, and the vertical whitespace and margins can withstand screen cropping.
When institution, date, address, price, discount, and similar details required for events, exhibitions, or promotions do not exist, leave blank space rather than filling the layout with fake information; non-event themes are not required to include an event information area.

Additional standards for square format

Under the default 1:1, the core remains prominent in a frame pressed from all four sides at once; a landscape or portrait composition must not simply be cropped into a square.
The subject, key numbers, or title are readable at a glance in a small feed; the four corners carry tension, rhythm, or whitespace, and do not pile up information of equal weight.
The background is fully opaque and forms a cohesive communication visual with the central focal point, text, or product.