Alain-lan/simple-sketch-image

Simple Sketch Image

Core Purpose

Create simple sketch illustrations that explain text clearly. Prefer white background, black hand-drawn line art, strong whitespace, and a single visible idea per image.

Select characters by explicit user intent first. If the user asks for no character, do not use one. If the user provides a character, brand mascot, reference image, or role description, use that as the fixed character. If the user names a built-in character, use it. If the user gives no character preference, prefer the default 线框人 only when a character helps explain the idea; pure object, icon, map, or process images may use no character.

Load References

Read only the files needed for the current request:

• references/visual-principles.md: style rules, default character, Chinese text limits, color rules, and forbidden styles.
• references/composition-types.md: choose image type for covers, article body illustrations, tutorials, workflows, comparisons, metaphors, and explanations.
• references/prompt-template.md: generation and editing prompt templates.
• references/qa-checklist.md: post-generation review and iteration rules.
• references/examples.md: concise examples for body illustration, cover, and tutorial prompts.

Reference selection:

• Read visual-principles.md for every image or prompt request.
• Read composition-types.md when choosing between cover, body illustration, tutorial, workflow, comparison, metaphor, route, or comic formats.
• Read prompt-template.md when writing generation prompts, editing prompts, or shot lists.
• Read qa-checklist.md after generating or reviewing an image.
• Read examples.md only when the request is ambiguous, the output format is unclear, or a concrete pattern would help.

Workflow

1. Understand the Source

Read the user's paragraph, article, outline, screenshot text, Markdown file, or brief. Extract:

• the core claim or main logic
• the section that most needs a visual anchor
• the intended use: cover, body illustration, tutorial step, explanation, metaphor, or visual note
• whether a fixed character is requested, optional, or forbidden
• whether Chinese labels are needed

For detailed character selection rules, use references/visual-principles.md.

Do not illustrate every paragraph by default. Choose only the strongest visual anchor unless the user asks for multiple images.

2. Choose the Output Mode

If the user asks for planning, return a concise shot list. For each image include:

• placement or use
• core meaning
• composition type
• character rule
• suggested elements
• suggested Chinese labels

If the user asks to generate images and an image generation tool is available, generate each image separately. If image generation is not available, output a complete copy-ready prompt for another image model.

3. Separate Cover and Body Rules

For a cover image:

• make the visual premise clear at first glance
• use one strong central metaphor or scene
• allow more visual contrast than body illustrations
• keep labels very short, or omit labels if the cover should carry external title text

For a body illustration:

• prioritize explanation over impact
• keep the drawing quieter and more spacious
• use 3-6 short labels only when they improve comprehension
• avoid turning the image into a slide, chart, or dense infographic

4. Generate or Write the Prompt

Use references/prompt-template.md for prompt construction. Include:

• aspect ratio, usually 16:9 horizontal
• white background
• black simple hand-drawn line art
• optional sparse accent colors
• character instructions if relevant
• one core idea
• concrete composition
• strict Chinese label limit
• negative constraints

5. Review and Iterate

After image generation, use references/qa-checklist.md. Regenerate or edit if:

• Chinese text is garbled or too long
• the image does not explain the source text
• the image is too complex
• it looks like a PPT slide, commercial vector illustration, 3D render, photo, or UI mockup
• the fixed character is missing, inconsistent, or decorative when the user requested it

Delivery

When generating images, return:

• how many images were generated
• what each image is for
• whether the text passed QA
• which image is strongest and which should be regenerated if needed
• saved path if files are created

When only producing prompts, return the prompt and a short note on how to use it.