Compile a title + purpose intoonedeterministic Swiss International Typographic Style poster prompt that can be handed directly to an image model.
Core principles:Options exist only in the reasoning and never appear in the output. Do not give the user a misleading menu of the "red/blue/yellow" kind—internally choose each module by intent, and output a single flat prompt with all parsing resolved.
First read
Read in the following order, then begin compiling:
references/compile-contract.md: Compilation contract (the 5 ironclad rules for disambiguation + the locked style core + fixed field order).Must be satisfied before every output.references/module-library.md: Module library (anchor / key visual / color scheme / title placement / aspect ratio / text mode) and the intent → module mapping table.references/presets.md: 6 compiled presets. When the user just wants something "ready to use" or does not want to describe it, directly give the closest set.
Input
Accept only two items; infer everything else:
- Title(required)—a Chinese theme word or a short title sentence. If missing, ask only this one question.
- Purpose—"display image / portfolio" or "communication cover". If unspecified, infer from the title's tone, and state the assumption in the selection rationale.
Do not ask the user to fill out a form, and do not show module options. If the user proactively specifies colors, aspect ratio, anchor designer, etc., the user's choice prevails (overriding inference).
Workflow
- Parse intent: Infer from the title and context the content type (methodology/tool/business/design/decision...), tone (rational and credible/high-energy warning), platform, and aspect ratio.
- Select modules: According to
module-library.md's mapping table, for each slot (anchor · key visual · color scheme · title placement · aspect ratio · text mode)assign exactly one value. When unsure, choose the best option; do not list alternatives side by side. - Compile: According to
compile-contract.md's field order and the 5 ironclad rules, assemble into one flat prompt—parse away all options, zero Chinese annotations, negatives only in AVOID, and declare the text to be rendered only once. - Output: One finished prompt + one line of "selection rationale" (explaining what purpose/content type was inferred, and which module set was chosen) + one line pointing to the closest preset (alternative).
Intent → Module Mapping (deterministic judgment layer)
See the detailed table in references/module-library.md. Key points:
- Purpose: Display image/portfolio → text mode TC (Latin main title + Chinese subtitle) or TA (pure Latin); communication cover → TB (Chinese main title + Latin small text, Chinese must be readable).
- Content type: Methodology/System/Knowledge → pure grid or color bands; Tools/Technology/AI → concentric arcs or rotating square matrix; Business/Report/Data → pure grid or color bands; Design/Type/Typography → type as image or rotating square matrix; Design experiment/Culture → controlled rule-breaking; Decision-making/Hierarchy → color bands.
- Tone: Rational and credible → cobalt blue / pure black and white; High-energy warning → yellow-black / red.
- Anchor follows the main visual: Arc/Grid→Müller-Brockmann, Square matrix/Color bands→Max Bill, Typography→Emil Ruder, Single large form/Black-and-white form→Armin Hofmann, Rule-breaking→Wolfgang Weingart.
Output format
Except for "ask a question when the title is missing," the final output:
- One complete, directly copyable English skeleton prompt (title declared with quotes/「」, conforming to the compilation contract).
- One line of rationale for the selection.
- One line: the closest preset number (from
presets.md), for the user to switch to directly if they want something more stable.
Do not output module menus, multi-version options, parameter tables, reasoning processes, or self-check records.
Quality threshold
- In the output,zero slash options, zero curly braces, zero Chinese annotations(except for the title text).
- One main visual, do not place two side by side.
- Negative words only in the AVOID block; the main text is entirely positive.
- The style core (unified grid / single grotesk / flat hard-edged with no gradients, 3D, or textures / objective with no decoration / asymmetric tendency) must be included in every item.
- Do not fabricate nonexistent image generation results; this skill only produces prompts.