Rewrite an article or a viewpoint into a four-panel comic that makes readers smile knowingly. The default style is black-and-white hand-drawn, 2x2 panels, xkcd-style deadpan humor structure, but the characters are not stick figures; they are uniformly replaced with panda meme characters that have a stronger Jin Guanzhang panda emoji flavor.
Default Goal
- Output one 2x2 four-panel comic PNG.
- The image should look like a black-and-white hand-drawn comic: black-and-white line art, small desktop props, exaggerated panda faces, Chinese dialogue, strong meme feel.
- Clearly explain the article's core expression, prioritizing funny points such as "contrast, misunderstanding, overpromising, confirm before dangerous actions, tools suddenly turning into butlers".
- The comic is suitable for insertion into the body of WeChat Official Account/Obsidian articles.
Style Anchors
Prefer using the reference image included with this skill as the style reference:
assets/reference-codex-computer-housekeeper-panda-comic.png
If the current environment supports viewing local images, first use view_image to open this reference image, then use it as the "Reference image: Jin Guanzhang panda four-panel comic style". If the image generation tool supports reference image input, use this image as the style reference; if it can only generate from text, write the reference image features into the prompt:
- Black-and-white line art, 2x2 four panels, clear comic borders.
- Panda characters with exaggerated real-person meme expressions: holding back laughter, shock, evil grin, sudden realization.
- Each panel has short dialogue bubbles and a few UI/desktop props.
- Medium image density, with everyday desktop details, avoiding pure icon infographics.
- The text is in Chinese, use short sentences as much as possible to reduce the risk of garbled generation.
Quality Baseline
- The final image should approach the level of completion of the reference image: obvious real-person meme face expressions, panda bodies with black-and-white fur texture, rich props such as desktops/computers/reports/interfaces, and overall like a black-and-white comic that can be posted directly to a WeChat Official Account.
- It is forbidden to deliver simple geometric pandas, flowcharts, small icons, or low-density wireframes as the final output.
- Each panel must have a specific scene and character interaction; it cannot just display a few isolated elements.
- Chinese dialogue should be short and precise first; small interface text can be simplified, but bubble text must be clear and readable.
- Only write natural dialogue in the bubbles, do not add speaker labels, and do not use the "character name: dialogue" format.
- Do not add periods at the end of bubble dialogue sentences; do not add "." to every sentence just to make it look complete; question marks and exclamation marks are only used when there is a genuine need to express doubt or strong emotion.
- If the local script generation effect does not reach the reference image standard, it should be redone using an image generation tool or a reference-image-driven approach; it can only be inserted into the article after viewing and confirmation.
Input Handling
Users usually provide:
- A Markdown article path, such as
1-Wechat/ing/xxx.md - The full text of an article
- A link or topic
- An already generated comic, hoping to consolidate it into a fixed style
First read or obtain the complete content. For a Markdown article, at least read the title, beginning, main subheadings, and ending; if the article is not long, read the full text.
Creation workflow
-
Distill one core statement
- Use 1 sentence to explain what the article truly wants readers to remember.
- Find the contrast best suited for comics: human expectations vs. the tool's actual behavior, old methods vs. new methods, surface function vs. real value.
-
Design a four-panel rhythm
- Panel 1: Set up the pain point or misunderstanding.
- Panel 2: The protagonist tries an ordinary approach, exposing clumsiness or absurdity.
- Panel 3: The panda character gives a smart but restrained solution action.
- Panel 4: Twist and close, stating the article's core viewpoint.
-
Write the storyboard table
- Each panel includes: visuals, character expressions, dialogue, hidden punchline.
- Keep each speech bubble within 12-22 Chinese characters.
- At most 2 speech bubbles per panel.
- Dialogue should sound like people talking, with less manual-style language.
- Write dialogue content directly, removing speaker labels and colon prefixes.
- By default, do not add a period at the end of dialogue; short sentences can stop at the text itself.
-
Generate images
- By default, call the available image generation tool to generate PNG, prioritizing black-and-white comic completeness at the level of the reference image.
- If the system has
imagegenskill orimage_gentool, use it. - Only when a local drawing script can reach the reference image's level of completeness may a script be used to generate the final image; otherwise the script may only be used for composition drafting and text repair.
- Save output preferentially to the article's sibling resource directory:
文章目录/assets/<文章slug>-panda-comic/<文章slug>-four-panel-panda-comic.png
- If there is no article path, save to:
output/xkcd-panda-comic/YYYY-MM-DD-主题/<主题>-four-panel-panda-comic.png
-
Quality check
- After generation, check with an image viewing tool.
- Confirm it is a 2x2 four-panel.
- Confirm the character is a panda meme expression and must not become a stick figure.
- Confirm the Chinese dialogue is generally readable.
- Confirm the fourth panel lands on the article's core viewpoint.
-
Handle Chinese garbled text
- If the Chinese text is obviously garbled, first regenerate once with shorter dialogue.
- If it is still garbled, keep the generated image's expressions and composition while creating an SVG/HTML/PNG text-corrected version, or provide overlayable Chinese dialogue in the final reply.
- When the user explicitly wants a black-and-white hand-drawn feel, prioritize preserving the generated image's meme temperament.
-
Naming and reporting wording
- Do not call the image or article a trial work, and do not use versioned wording that implies "not yet delivered."
- If the image was retouched or redone, only say "generated," "corrected," or "updated," and explain the final result.
- Use stable finished-product naming for file names, such as
four-panel-panda-comic.png, avoiding English experimental version markers.
Image generation Prompt template
Replace the bracketed content in the template below with the current article information:
Use case: illustration-story
Asset type: 2x2 four-panel comic for a Chinese tech article
Primary request: Turn the article's core idea into a funny four-panel comic. Core idea: [一句核心表达]
Style/medium: black-and-white hand-drawn webcomic, xkcd-like dry humor and simple panel rhythm, but no stick figures. Use panda meme characters with a strong Chinese “金馆长熊猫表情 / 金教授熊猫脸” vibe: round panda heads, black ears, exaggerated human-like meme facial expressions, smug grin, shocked crying face, sly laugh, and sudden enlightenment.
Composition/framing: 2x2 panel grid, clear panel borders, each panel has one joke beat. Add speech bubbles with short clean Chinese text. Include small props related to the article: [道具列表].
Characters:
- Main human/user: panda at a messy desk, emotionally dramatic.
- Smart tool/agent: confident panda helper with clipboard or laptop, calm and slightly smug.
- Optional old-method foil: overdramatic panda selling scary one-click optimization.
Panel script:
1. [第一格画面和对白]
2. [第二格画面和对白]
3. [第三格画面和对白]
4. [第四格画面和对白]
Text (verbatim, Chinese; no sentence-final full stop in speech bubbles):
Title: “[短标题]”
Panel 1: “[对白 1,不带句末句号]” “[对白 2,不带句末句号]”
Panel 2: “[对白 1,不带句末句号]” “[对白 2,不带句末句号]”
Panel 3: “[对白 1,不带句末句号]” “[对白 2,不带句末句号]”
Panel 4: “[对白 1,不带句末句号]” “[对白 2,不带句末句号]”
Constraints: Chinese text should be clean and legible; speech bubbles should contain direct dialogue only, with no speaker-name prefix or name-colon format; do not add Chinese full stops “。” at the end of speech-bubble sentences; keep sentences short; black, white, and light gray only; no watermark; no logo; no photorealistic scene; no stick figures; no colorful cute cartoon; no dense paragraphs.
Avoid: malformed Chinese text, extra panels, brand logos, realistic celebrity likeness, horror, violent imagery, insulting real people.
Storyboard writing techniques
- Prioritize turning the article's "feature list" into character actions; do not stuff the list into the frame as-is.
- Old tools can handle exaggeration and misdirection, while new tools handle evidence, process, and restraint.
- The fourth panel is best with one line that can be screenshotted and shared on its own.
- Common punchlines in technical articles:
- "I only wanted to clear a cache, and you installed a whole bundle for me"
- “A truly reliable butler gathers evidence first, then acts”
- “The stronger the tool, the more you must know to stop and wait for confirmation”
- “Less than 1% developed, and it’s already managing my computer”
Delivery format
After completion, reply to the user:
老板,已生成四格漫画:
- 图片:[绝对路径]
- 风格:金馆长熊猫表情 / 黑白四格
- 核心梗:[一句话]
- 是否已插入文章:[是/否]
可插入 Markdown:

If modified .md file and it contains non- r2blog.zhanglearning.com / r2.zhanglearning.com of http(s) external image links, call at the wrap-up stage 1-upload-images-to-picgo to migrate the image hosting. When only generating local PNGs and inserting them with relative paths, no migration is needed.
Example
Input:
我有文章,需要生成 xkcd 风格四格漫画,人物换成金馆长熊猫表情。
文章:1-Wechat/ing/2026-06-23-Codex 是最好的电脑管家.md
Output approach:
核心梗:传统电脑管家先吓人,Codex 先查证、列风险、等确认。
四格:
1. 老板电脑空间红了,传统管家跳出来喊“99999 个风险,先装全家桶”。
2. 老板让 Codex 清理 uv 缓存,Codex 先统计目录和运行任务。
3. Codex 递出清理报告:可删 2.5GiB,危险动作等确认。
4. Codex 展示卸载残留、启动项、大文件、巡检,老板感叹“我只开发了不到 1%”。