WeChat Doodle Illustrations

Overview

Generate finished WeChat article illustrations by default, not just prompts. Keep the style like a scanned notebook page: textured white ruled paper, pale blue horizontal lines, one red margin line, thick black marker doodles, rough hand-lettered Chinese text, imperfect boxes, tiny marginal notes, simple icons, and one or two small color accents.

The target should feel like a real marker sketch scanned from a notebook, not a clean SVG, poster, UI mockup, or digital illustration. All visible text must look hand-drawn: uneven brush pressure, irregular character size, slightly crooked baseline, and no clean digital fonts.

Load references/style-guide.md before writing final prompts or generating images.

Style Baseline

Use assets/examples/ as the current visual benchmark. New outputs should match these examples more than a clean diagram:

• assets/examples/codex-mobile-cover.png for cover density, paper texture, hand-lettered headline scale, and 2.35:1 crop safety.
• assets/examples/codex-mobile-work-anywhere.png, assets/examples/codex-mobile-decision.png, and assets/examples/codex-mobile-remote-ssh.png for in-article 16:9 composition, rough UI cards, thick marker linework, and small red accent usage.

Treat assets/style-reference.jpg as a general style reference, and assets/examples/ as the practical production standard for this skill.

Workflow

1. Read the article or brief and extract the core reader promise in one sentence.
2. Split the article into 3-6 visual beats: cover, key concept, process, contrast, example, ending note.
3. Choose the minimum useful image set:
• For a cover: create one strong 2.35:1 composition.
• For a full article: create 1 cover plus 3-5 in-article illustrations.
• For ideation: return a shot list only when the user asks for ideas or prompts.
4. Use concrete metaphors instead of abstract AI imagery. Prefer rough tool panels, notes, windows, arrows, checklists, robots, cursor pointers, folders, small people, and messy annotations.
5. Make each image readable on mobile. Use 1 focal idea, 3-7 words of visible text at most, and avoid dense paragraph text inside the image.
6. When the image contains Chinese text, explicitly request bold hand-lettered Chinese marker characters, not typed font, not calligraphy, not printed typography.
7. For a full article visual package, generate the cover first and wait for style confirmation unless the user explicitly asks to generate the whole set in one pass.
8. Generate finished images directly when an image generation tool is available.
9. Include a negative prompt internally that blocks clean digital fonts, polished vector art, 3D render, glossy gradients, corporate stock illustration, photorealism, and overly cute mascot style.

Prompt Quality Lock

Direct image generation must still be prompt-first. Before calling any image generation tool, write a complete final English prompt internally using the same level of detail as the prompt-only output. Do not replace it with a short summary.

The prompt must include:

• Hand-drawn notebook doodle illustration on white ruled paper with pale blue horizontal lines and one red margin line
• black marker pen linework
• slightly imperfect rectangles and arrows
• bold hand-lettered Chinese headline
• sparse layout, mobile-readable
• one small color accent only
• scanned paper texture, natural shadows, high contrast black ink

For Chinese text, prefer this wording:

bold hand-lettered Chinese headline "[exact text]", rough black marker strokes, uneven pressure, irregular character size, slightly crooked baseline, not a typed font

Do not let direct generation simplify the style into clean UI art, SVG-like illustration, or system-font typography.

Output Format

Default output is finished images plus a short usage note. Do not expose long prompts unless the user asks for them.

If the user asks for 提示词, Prompt, prompt-only, 只要提示词, or 不要出图, return prompt-only output instead of generating images.

For direct image generation:

• First create the complete final prompt internally using Prompt Quality Lock.
• Generate the image with the image generation tool.
• Show the finished image in the response when possible.
• Add a short note with 用途, 建议位置, and 尺寸.
• Keep the written explanation concise.

For prompt-only requests, provide:

• 用途: cover / section / transition / ending
• 画面: one-sentence composition
• 文字: exact visible words, if any
• Prompt: final English image prompt
• Negative prompt: style exclusions
• 尺寸: recommended ratio

Use assets/style-reference.jpg as the local style reference when the image generation tool supports reference images.

Direct Image Generation Rules

• If the user says 直接出图, 生成图片, 成品图, 配图, 封面图, or gives an article and asks for visuals, generate finished images directly.
• If the user asks for prompts instead of finished images, do not call the image generation tool.
• If the user asks for a full article visual package, start with 1 cover image first and ask for style confirmation before generating in-article illustrations, unless they explicitly ask for all images at once.
• If generating multiple images, keep the set visually consistent: same paper texture, same black marker line weight, same accent color logic.
• If text accuracy inside the image matters, keep visible text very short and repeat the exact Chinese words in the prompt.
• If style accuracy matters, prioritize hand-lettering, rough scanned paper texture, and imperfect marker linework over perfect text layout.
• Do not prescribe a default output folder. Save files only when the user or host project provides an output location; otherwise return the generated images in the response.
• If the tool cannot generate images, fall back to prompt-only output and clearly say so.

Preflight Checklist

Before generating or returning prompts, quickly verify:

• Cover images use 2.35:1; in-article images use 16:9.
• Full article packages start with 1 cover for style confirmation unless the user asks for all images at once.
• The prompt aligns with assets/examples/ for paper texture, rough black marker linework, hand-lettered Chinese titles, hand-drawn UI cards, and restrained red accents.
• Visible text is short, mobile-readable, and described as hand-lettered marker text rather than typed typography.
• The negative prompt blocks SVG-like art, clean digital fonts, polished vector illustration, 3D render, corporate stock illustration, and dense infographics.

WeChat Defaults

• Cover ratio: 2.35:1; keep the main subject centered because WeChat crops previews.
• In-article ratio: 16:9; keep line art thick enough for phone screens.
• Text language: default to Chinese unless the article title or user brief is English.
• Tone: clever, handmade, slightly messy, designer-notebook energy; avoid childish classroom worksheet vibes.
• Typography: visible words should look written by hand with a fat black marker. Avoid system fonts, clean sans-serif, vector lettering, perfect alignment, or poster-style typography.

Prompt Skeleton

Hand-drawn notebook doodle illustration on textured white ruled paper with pale blue horizontal lines and one red margin line, rough black marker pen linework with uneven pressure, slightly imperfect rectangles and arrows, tiny handwritten annotations, simple tool icons and small doodle characters, [subject/composition], [visible text as bold hand-lettered Chinese marker characters, irregular baseline, not typed font], sparse layout, mobile-readable, editorial illustration for a WeChat article, one small color accent only, scanned paper texture, natural shadows, high contrast black ink.