Compile the user's given theme, article, title or emotion intooneexperimental poster prompt that can be handed directly to an image model. The core of the style is the spatial structure of "clear foreground layer + blurred background layer", with the vibe landing between experimental magazine cover / youth culture poster / underground flyer / Y2K / Acid Graphics.
Core principles:The template body is locked down; the input area is open. All of the skill's work lies in filling in the input area well; the body is copied verbatim without changing a single character, leaving creative freedom to the image model.
Read
- Generate prompt: read
references/master-template.md(the single source of truth template).
Input
Infer from user input first; do not require filling in a form. The input area has 12 fields in total, and only "Theme" is required:
| Field | Filling judgment |
|---|---|
| Theme | Required. Distill from the article, title or user description into one sentence, with an emotional direction, not a dry category word |
| Main title | If the user gave a title, copy it verbatim without changing a single character; if not given and a strong title can be proposed from the article, fill in a short title of 2-8 characters; if not confident, leave blank |
| Subtitle | Fill in if there is one ready-made; otherwise leave blank |
| Title styling tendency | Choose 2-3 from the template's [main title glyph styling] technique list, matched according to the theme's temperament (restless→broken strokes/spray paint/extrusion collision; nostalgic→old print/brush/abrasion; restrained→extreme stretch/sharp cut corners); a single image may leave it blank to let the model choose,when in a group it is required and each image must not repeat the others |
| Core keywords | Propose 2-4 from the theme to help the image model grasp the atmosphere; leave blank if unsure |
| Auxiliary text | Fill in only if there are concrete short sentences (opinion sentences, dates, numbers); do not make things up just to fill it in |
| English short sentences or tags | Fill in only if there is something that naturally fits (such as product name, year, slogan); do not force decorative English |
| Color tendency | Fill in according to the spatial mode (see Workflow step 2), format: mode + single foreground spot color + background field/photo color temperature. When the user specifies colors, fit them into the mode format; leave blank only if completely unsure |
| Image atmosphere | Fill in when a clear emotion can be read from the theme (such as: the restlessness of rushing drafts late at night, the excitement before release); leave blank if vague |
| Photo or scene tendency | Fill in when the theme has a concrete scene anchor, and try to write out three layers of depth of field: foreground silhouette, midground subject, background blurred light source; otherwise leave blank |
| Aspect ratio | Prefer user specification; infer by use: X/WeChat official account cover horizontal 16:9 or 5:2, poster vertical 3:4 or 2:3; if there is no clue at all, leave blank |
| Text language | Default to mainly Chinese; change only if the user wants an English version |
Leaving blank is a legal output, not laziness: the template clearly stipulates that blank fields are left to the image model's free judgment.Better to leave blank than to fabricate——Filling in a wrong specific value (a fake date, forced English, unfounded color scheme) hurts the image more than leaving it blank.
Workflow
- Identify input: Determine whether it is a single generation, a group generation, or the user just wants a blank template.
- Choose spatial mode: This is the key decision for producing the "naked-eye 3D" sense of depth, choose one of two——
- Mode one (photographic base): Suitable for themes with a specific sense of scene (last bus, convenience store, internet cafe, rooftop). Dark base + depth photo retaining its own color temperature, foreground spot color in warm-cold opposition to the photo's color temperature.
- Mode two (print base): Suitable for opinion-type, manifesto-type, brand-type themes (inaugural issue, account poster, attitude manifesto). Full-page high-saturation solid color base + photo dyed into the base color family and heavily blurred, foreground spot color in high-purity clashing vibration with the base color. The foreground always uses onlyonespot color flat fill, with layering relying on changes in font size and form, not on switching colors.
- Distill the theme: Compress the article or description into one emotional theme sentence. If the user gives exact text (title, subtitle, short phrase), it must be kept verbatim, not translated, not polished.
- Fill the input area: Judge fill or leave blank field by field according to the table above. Write "color scheme tendency" as a structured description according to the chosen mode; for "photo or scene tendency" try to give three layers: close shot/medium shot/long shot.
- Assemble output: The filled input area + filling instructions + template fixed body text (from 【生成任务】 to 【最终效果】, copied verbatim) + exclusion section (attached by default; omitted when the user says "simplified version"). Put the whole thing into a ```text code block, ready to copy and use.
- Group generation: Produce one complete prompt for each image separately, with the input area varying at least two of "color scheme tendency, image atmosphere, photo or scene tendency," and both spatial modes must appear, to avoid homogenization across the whole group.
Output Format
## 已填输入区
- 主题：…
- 主标题：…（留空字段标「留空→模型自判」）
- 画幅比例：…
## 完整提示词
```text
【用户输入区】
主题：…
主标题：…
（11 个字段全部列出，留空的字段保留字段名、值为空）
填写说明：
除“主题”外，其余内容均可留空。留空的部分由模型根据主题和整体视觉方向自由判断与补充。
【生成任务】
…（固定正文原样到【最终效果】结束）
Avoid: every element sitting at the same focus depth; …（原样照抄 master-template.md 的排除项段）
```
## 填法理由
一两句：主标题为什么这样定、场景/配色为什么填或留空。
When the user only wants the prompt, omit "filled input area" and "filling rationale" and give only the code block. When the user wants a blank template, directly output the input area in master-template.md + fixed body text + exclusion section, without any filling.
Quality Gates
- The fixed body text must be exactly consistent with
master-template.md: do not delete sections, do not abbreviate, do not change wording, do not adjust paragraph order. The template body text itself is a designed prompt, and any change degrades it. - "Color scheme tendency" must reflect a single foreground spot color + a clear mode attribution; any filling that has "two or three foreground colors side by side" or "foreground color in the same color family as the photo" is a failure.
- The photo description for mode one must include warm-cold opposition; the photo description for mode two must state "dyed into the background color family, heavily blurred, blended into the base color."
- The main title is a "constructed title lettering," not standard print type set straight; in group output, the combinations of letterform styling techniques for each image must not repeat.
- User-specified text (main title, subtitle, auxiliary text) must appear verbatim in the input area, not translated, not altered.
- Do not fabricate specific values for blank fields; do not generate fake dates, fake locations, or fake numbers (unless the theme truly includes them).
- The output must be one complete independent prompt; the user can generate an image by copying the code block, without needing to look at this skill again or add anything.
- In group output, every prompt is complete and self-sufficient, and the input areas have clear differences.
- This skill only produces prompts and does not fabricate image results that have already been generated.