punk-cover

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Punk Cover

Punk 封面

Core Rule

核心规则

punk-cover

is a cover-generation workflow, not a style owner. Select exactly one reusable style from the repository-level

styles/

library, then compile that style atom into the cover shape.

The final image prompt is not a loose concatenation of "cover instructions + style instructions". It must be one integrated, cover-specific prompt that applies the selected style to the cover format.

Use three inputs:

```
punk-cover
```
task fields: platform, aspect ratio, title clarity, article summarization, cover communication goals, and universal output constraints.
The selected style's
```
META.md
```
metadata and
```
STYLE.md
```
reusable visual style atom.
```
references/cover-prompt-blueprint.md
```
, which defines the full cover prompt shape.

The style file defines the reusable visual language. The cover blueprint defines the output form. The final prompt must fuse them into one complete cover-generation prompt.

punk-cover

是一个封面生成工作流，而非风格所有者。需从仓库级别的

styles/

库中精准选择一种可复用风格，再将该风格原子整合到封面形态中。

最终的图像提示词不能是“封面说明 + 风格说明”的松散拼接，必须是将选定风格应用于封面格式的、一体化的专属封面提示词。

需使用三类输入：

```
punk-cover
```
任务字段：平台、宽高比、标题清晰度、文章摘要、封面传播目标以及通用输出约束。
选定风格的
```
META.md
```
元数据和
```
STYLE.md
```
可复用视觉风格原子。
```
references/cover-prompt-blueprint.md
```
，它定义了完整的封面提示词结构。

风格文件定义可复用的视觉语言，封面蓝图定义输出形式，最终提示词必须将二者融合为一个完整的封面生成提示词。

Resources

资源

Read
```
references/style-catalog.md
```
to list or choose cover styles.
Read
```
references/cover-prompt-blueprint.md
```
before composing the final prompt.

For the selected style, read both:

```
../../styles/{style-id}/META.md
```
```
../../styles/{style-id}/STYLE.md
```

Only expose styles whose
```
META.md
```
metadata has
```
outputs
```
containing
```
cover
```
or
```
poster
```
.
Do not expose photo, avatar, portrait, pet-only, polaroid-only, or image-remix styles in the
```
punk-cover
```
menu unless their metadata explicitly includes
```
cover
```
or
```
poster
```
.
Do not use legacy cover templates. Use only the selected style's
```
META.md
```
,
```
STYLE.md
```
, and
```
references/cover-prompt-blueprint.md
```
.

阅读
```
references/style-catalog.md
```
以列出或选择封面风格。
撰写最终提示词前，请先阅读
```
references/cover-prompt-blueprint.md
```
。

对于选定的风格，需同时阅读：

```
../../styles/{style-id}/META.md
```
```
../../styles/{style-id}/STYLE.md
```

仅展示其
```
META.md
```
元数据中
```
outputs
```
包含
```
cover
```
或
```
poster
```
的风格。
除非元数据明确包含
```
cover
```
或
```
poster
```
，否则在
```
punk-cover
```
菜单中不展示仅适用于照片、头像、肖像、宠物、拍立得或图像 remix 的风格。
请勿使用旧版封面模板，仅使用选定风格的
```
META.md
```
、
```
STYLE.md
```
以及
```
references/cover-prompt-blueprint.md
```
。

Workflow

工作流

Analyze the source material:
- Extract the core visual subject.
- Preserve the complete user title or topic when one exists, but do not copy the full article body into any output field.
- Draft an optional subtitle when the selected style has a subtitle field.
- Summarize context, audience, mood, metaphor, and banned elements.
- Convert long source material into concise derived fields: title/topic, 1-3 sentence context summary, visual subject, audience, mood, metaphor, and banned elements.
Confirm platform and aspect ratio before generating any prompt:
- Xiaohongshu:
```
3:4
```
- WeChat public account:
```
2.35:1
```
- X:
```
5:2
```
- Custom: keep the user's ratio exactly.
- If the user only provides source content, ask which platform they want to publish to. Do not generate the prompt yet.
- If the user provides a custom ratio, use it and do not ask for platform unless the platform matters for wording.
- Only skip this question when the user has already provided a platform, a ratio, or explicitly says to decide everything automatically.
Confirm style before generating any prompt:
- If the user specifies one catalog style, use it.
- If the user supplies a complete visual direction that matches a catalog style, including the
```
复古时代错位编辑封面
```
  brief, treat that style as specified and use the matching
```
META.md
```
  and
```
STYLE.md
```
  without asking the user to repeat the style.
- If no style is specified, recommend exactly three eligible catalog styles based on the content and give a one-sentence reason for each, then ask the user to choose one or provide a custom style direction.
- Do not show all eligible styles by default unless the user asks for the full menu.
- Only auto-select one style when the user explicitly says to decide everything automatically, not merely because they provided an article.
Use this confirmation gate:
- When platform/ratio or style is missing, stop after asking the question. Do not fill a style file, save prompt files, or generate an image.
- The first response for article-only input should contain the platform question and three recommended styles.
- If both platform/ratio and style are known, continue without asking.
Compile the final image prompt:
- Use
```
references/cover-prompt-blueprint.md
```
  as the required structure for
```
prompts/cover.md
```
  .
- Read exactly one selected style visual atom from
```
../../styles/{style-id}/STYLE.md
```
  .
- Read the selected style metadata from
```
../../styles/{style-id}/META.md
```
  , including
```
style_anchors
```
  ,
```
cover_shape_adaptation
```
  ,
```
must_preserve
```
  , and
```
avoid_when_applying_to_cover
```
  when present.
- Extract the selected style's non-negotiable visual anchors: materials, spatial logic, title treatment, typography behavior, texture, color system, and style-specific negative constraints.
- Fuse the cover task fields and style anchors into one full cover prompt. The final prompt should read like a complete cover-generation brief, similar to the legacy complete prompts in
```
exports/
```
  , not like two unrelated prompt fragments pasted together.
- Include cover-shape sections for role/task, input fields, content understanding, title hierarchy, cover composition, style application, image-text relationship, typography, color/material/texture, negative constraints, and final standard.
- Every cover decision must be expressed through the selected style. For example, if the style is torn collage, the title, subject, support text, background, and metaphor must be implemented as torn paper, old newspaper, tape, grain, halftone, and paper layers; not as a generic cover with collage words appended.
- Do not append the raw
```
STYLE.md
```
  content as a standalone second section. Rewrite and adapt its style atoms into the cover blueprint.
- Do not combine multiple styles or add a second custom style section.
- Replace or resolve all explicit placeholders such as
```
{{主题词}}
```
  ,
```
{{副标题，可留空}}
```
  ,
```
{{画幅比例...}}
```
  ,
```
{{语言...}}
```
  ,
```
{{用途...}}
```
  ,
```
{{补充背景，可留空}}
```
  ,
```
{{情绪倾向...}}
```
  ,
```
{{不想出现的元素，可留空}}
```
  , and any other
```
{{...}}
```
  fields.
- If a needed detail has no matching placeholder, merge it into the nearest cover section such as
```
补充语境
```
  ,
```
风格落地方式
```
  , or
```
禁用元素
```
  .
- For long articles, fill title/topic fields with concise derived titles or title layers, not the article text.
- Put only summarized context into the prompt; do not paste the original article body into the final prompt.
- Leave optional fields blank only when the prompt says they can be blank.
- Do not output analysis inside the final prompt.
Save files before image generation:
- Create
```
punk-assets/punk-cover/{slug}/prompts/cover.md
```
  with the complete filled prompt.
- If image generation returns an explicit local file path, downloadable URL, or image bytes for the current run, save that artifact as
```
punk-assets/punk-cover/{slug}/cover.png
```
  .
- Do not infer the correct artifact by scanning broad generated-image directories, because those directories may contain unrelated images from other runs.
- If the image-generation tool only returns an inline preview with no explicit retrievable artifact for the current run, do not create a fake
```
cover.png
```
  .
Generate an image by default after saving the prompt when a usable image-generation tool is available, such as
```
image_gen
```
. Treat inline preview generation as successful image generation, but treat local
```
cover.png
```
saving as successful only when the tool explicitly exposes a current-run artifact. Skip image generation only when the user explicitly asks for prompt-only output or the current environment has no image-generation tool. If image generation is unavailable, return the prompt file path and the full prompt content.

分析源素材：
- 提取核心视觉主题。
- 若存在完整的用户标题或主题，请予以保留，但请勿将整篇文章内容复制到任何输出字段中。
- 当选定风格包含副标题字段时，可草拟可选副标题。
- 总结背景信息、受众群体、情绪氛围、隐喻意象以及禁用元素。
- 将长篇源素材转换为简洁的衍生字段：标题/主题、1-3句背景摘要、视觉主题、受众、情绪、隐喻、禁用元素。
生成提示词前确认平台和宽高比：
- 小红书：
```
3:4
```
- 微信公众号：
```
2.35:1
```
- X：
```
5:2
```
- 自定义：严格保留用户提供的比例。
- 若用户仅提供源内容，请询问其发布平台，暂不生成提示词。
- 若用户提供自定义比例，则使用该比例，除非平台会影响措辞，否则无需询问平台。
- 仅当用户已明确提供平台、比例或要求自动决定所有事项时，才可跳过此步骤。
生成提示词前确认风格：
- 若用户指定了某一目录风格，则使用该风格。
- 若用户提供的完整视觉方向与某一目录风格匹配（包括
```
复古时代错位编辑封面
```
  的简要描述），则将该风格视为指定风格，直接使用对应的
```
META.md
```
  和
```
STYLE.md
```
  ，无需让用户重复说明风格。
- 若未指定风格，则根据内容精准推荐三种符合条件的目录风格，并为每种风格附上一句适配内容的推荐理由，然后请用户选择其中一种或提供自定义风格方向。
- 默认情况下请勿展示所有符合条件的风格，除非用户要求查看完整菜单。
- 仅当用户明确要求自动决定所有事项时，才可自动选择一种风格，不能仅因用户提供了文章就自动选择。
使用确认关卡：
- 若缺失平台/比例或风格，询问后停止操作，请勿填充风格文件、保存提示词文件或生成图像。
- 仅输入文章的情况下，首次回复应包含平台询问和三种推荐风格。
- 若平台/比例和风格均已知，则无需询问，继续操作。
编写最终图像提示词：
- 以
```
references/cover-prompt-blueprint.md
```
  作为
```
prompts/cover.md
```
  的必填结构。
- 从
```
../../styles/{style-id}/STYLE.md
```
  中精准读取选定的风格视觉原子。
- 读取选定风格的
```
META.md
```
  元数据，包括（若存在的话）
```
style_anchors
```
  、
```
cover_shape_adaptation
```
  、
```
must_preserve
```
  和
```
avoid_when_applying_to_cover
```
  。
- 提取选定风格中不可协商的视觉锚点：材质、空间逻辑、标题处理方式、排版表现、纹理、色彩系统以及风格专属的负面约束。
- 将封面任务字段和风格锚点融合为一个完整的封面提示词。最终提示词应读起来像一份完整的封面生成 brief，类似
```
exports/
```
  中的旧版完整提示词，而非两个无关提示片段的粘贴。
- 需包含封面形态的各个板块：角色/任务、输入字段、内容理解、标题层级、封面构图、风格应用、图文关系、排版、色彩/材质/纹理、负面约束以及最终标准。
- 所有封面决策必须通过选定风格来体现。例如，若风格是撕裂拼贴，则标题、主题、辅助文字、背景和隐喻都必须以撕纸、旧报纸、胶带、颗粒、半色调和纸张层的形式呈现；不能是一个通用封面再附加拼贴字样。
- 请勿将原始
```
STYLE.md
```
  内容作为独立的第二部分附加，需将其风格原子重写并适配到封面蓝图中。
- 请勿混合多种风格或添加第二个自定义风格板块。
- 替换或解析所有明确的占位符，例如
```
{{主题词}}
```
  、
```
{{副标题，可留空}}
```
  、
```
{{画幅比例...}}
```
  、
```
{{语言...}}
```
  、
```
{{用途...}}
```
  、
```
{{补充背景，可留空}}
```
  、
```
{{情绪倾向...}}
```
  、
```
{{不想出现的元素，可留空}}
```
  以及其他任何
```
{{...}}
```
  字段。
- 若所需细节没有匹配的占位符，则将其合并到最接近的封面板块中，例如
```
补充语境
```
  、
```
风格落地方式
```
  或
```
禁用元素
```
  。
- 对于长篇文章，标题/主题字段需填写简洁的衍生标题或标题层级，而非文章原文。
- 仅将总结后的背景信息放入提示词中，请勿将原始文章内容粘贴到最终提示词里。
- 仅当提示词明确说明可选字段可留空时，才将其留空。
- 请勿在最终提示词中包含分析内容。
生成图像前保存文件：
- 创建
```
punk-assets/punk-cover/{slug}/prompts/cover.md
```
  文件，填入完整的提示词内容。
- 若图像生成返回当前运行的明确本地文件路径、可下载URL或图像字节流，则将该产物保存为
```
punk-assets/punk-cover/{slug}/cover.png
```
  。
- 请勿通过扫描宽泛的生成图像目录来推断正确产物，因为这些目录可能包含其他运行产生的无关图像。
- 若图像生成工具仅返回内嵌预览图，未提供当前运行的可检索产物，则请勿创建虚假的
```
cover.png
```
  文件。
当有可用的图像生成工具（如
```
image_gen
```
）时，保存提示词后默认生成图像。将内嵌预览图生成视为图像生成成功，但仅当工具明确提供当前运行的产物时，才视为本地
```
cover.png
```
保存成功。仅当用户明确要求仅输出提示词，或当前环境无图像生成工具时，才跳过图像生成步骤。若无法生成图像，则返回提示词文件路径和完整的提示词内容。

First Response Format

首次回复格式

For article-only input with no platform and no style, ask concisely:

Which platform/aspect ratio should this cover target?
- Xiaohongshu:
```
3:4
```
- WeChat public account:
```
2.35:1
```
- X:
```
5:2
```
- Custom: provide the ratio
Recommended styles:
- ```
Style A
```
  : reason tied to the article.
- ```
Style B
```
  : reason tied to the article.
- ```
Style C
```
  : reason tied to the article.

End by asking the user to choose a platform and one style, or to say "auto" if they want the skill to decide everything.

对于仅输入文章、未指定平台和风格的情况，请简洁询问：

该封面应适配哪个平台/宽高比？
- 小红书：
```
3:4
```
- 微信公众号：
```
2.35:1
```
- X：
```
5:2
```
- 自定义：请提供比例
推荐风格：
- ```
Style A
```
  ：与文章适配的理由。
- ```
Style B
```
  ：与文章适配的理由。
- ```
Style C
```
  ：与文章适配的理由。

最后请用户选择平台和一种风格，或回复“auto”由该技能自动决定所有事项。

Style Selection Heuristics

风格选择启发式规则

Use styles whose
```
outputs
```
metadata contains
```
cover
```
or
```
poster
```
.
Use business/report styles for strategy, product, AI, startup, industry, consulting, or analysis content.
Use
```
复古时代错位编辑封面
```
for AI, coding, digital work, future tools, or contemporary topics that benefit from a human-centered mid-century illustration and one restrained era-displacement metaphor.
Use journal/concept styles for science, research, medicine, engineering, or mechanism-heavy content.
Use collage, giant-title, block, brick, or diffuse styles for social posts that need stronger shareability.
Use black-white minimal or avant-geometry styles for abstract, philosophical, critical, or high-contrast editorial themes.
Do not recommend styles whose metadata is avatar-only, portrait-only, pet-only, polaroid-only, or image-only unless they also declare
```
cover
```
or
```
poster
```
.

使用其
```
outputs
```
元数据包含
```
cover
```
或
```
poster
```
的风格。
针对战略、产品、AI、创业、行业、咨询或分析类内容，使用商务/报告风格。
针对AI、编程、数字工作、未来工具或适合以人为中心的中世纪插画搭配克制时代错位隐喻的当代主题，使用
```
复古时代错位编辑封面
```
风格。
针对科学、研究、医学、工程或侧重机制的内容，使用期刊/概念风格。
针对需要更强传播性的社交帖子，使用拼贴、大标题、块体、砖块或弥散风格。
针对抽象、哲学、批判性或高对比度的编辑主题，使用黑白极简或先锋几何风格。
除非同时声明了
```
cover
```
或
```
poster
```
，否则请勿推荐仅适用于头像、肖像、宠物、拍立得或纯图像的风格。

Cover Prompt Blueprint

封面提示词蓝图

Use

references/cover-prompt-blueprint.md

as the full prompt structure. The block below is retained only as the required task-field content that must be represented inside the integrated final prompt.

text

undefined

以

references/cover-prompt-blueprint.md

作为完整的提示词结构。下方的区块仅作为必须整合到最终提示词中的必填任务字段内容予以保留。

text

undefined

punk-cover cover task instructions

Create one single cover image for {platform}. Aspect ratio: {ratio}.

Use the following derived content only:

Title/topic: {title_or_topic}
Optional subtitle: {subtitle}
Short context summary: {summary}
Visual subject: {visual_subject}
Audience: {audience}
Mood: {mood}
Visual metaphor: {metaphor}
Banned elements: {banned_elements}

The main title must be complete, accurate, and clearly readable. If the source title is long, extract a short high-impact visual title while preserving the complete meaning through a smaller title, subtitle, or context line.

For long articles, use only derived fields such as title, summary, visual subject, metaphor, and supplemental context. Do not paste the original article body into the image, prompt, metadata, or small text system.

The cover must work for sharing: first glance identifies the topic, second glance reveals the visual metaphor. The image should feel like a deliberate editorial cover, not a generic illustration.

Avoid universal cover failures: PPT cover feel, course-cover feel, generic information-graphic template, e-commerce advertisement, unrelated decoration, misspelled title, missing title, title cropped beyond recognition, or title severely blocked by visual elements.

Generate only one final image. Do not output explanations, alternatives, grids, contact sheets, or multi-option compositions.

undefined

Create one single cover image for {platform}. Aspect ratio: {ratio}.

Use the following derived content only:

Title/topic: {title_or_topic}
Optional subtitle: {subtitle}
Short context summary: {summary}
Visual subject: {visual_subject}
Audience: {audience}
Mood: {mood}
Visual metaphor: {metaphor}
Banned elements: {banned_elements}

The cover must work for sharing: first glance identifies the topic, second glance reveals the visual metaphor. The image should feel like a deliberate editorial cover, not a generic illustration.

Generate only one final image. Do not output explanations, alternatives, grids, contact sheets, or multi-option compositions.

undefined

Output Discipline

输出规范

The final generated prompt must be one integrated cover-specific prompt compiled from the cover task fields,
```
references/cover-prompt-blueprint.md
```
, and exactly one selected style atom.
Do not paste the selected style
```
STYLE.md
```
as a standalone second section. Adapt its style atoms into the cover blueprint.
The final prompt must preserve the selected style's non-negotiable anchors from
```
META.md
```
and
```
STYLE.md
```
.
Do not copy long source text into the final prompt; use summaries and extracted fields.
Do not include style-selection rationale inside the prompt file.
Do not combine multiple styles.
Do not add a second custom style section beyond the selected style visual layer.
Do not expose non-cover styles in the style menu.

最终生成的提示词必须是整合了封面任务字段、
```
references/cover-prompt-blueprint.md
```
和恰好一种选定风格原子的一体化专属封面提示词。
请勿将选定风格的
```
STYLE.md
```
作为独立的第二部分粘贴，需将其风格原子适配到封面蓝图中。
最终提示词必须保留选定风格在
```
META.md
```
和
```
STYLE.md
```
中不可协商的锚点。
请勿将长篇源文本复制到最终提示词中，使用摘要和提取的字段即可。
请勿在提示词文件中包含风格选择的理由。
请勿混合多种风格。
除选定的风格视觉层外，请勿添加第二个自定义风格板块。
请勿在风格菜单中展示非封面类风格。