5km Littlebox Illustrations

Create clean, strange, hand-drawn article illustrations built around Littlebox (小盒): a small paper-box character that turns messy ideas into structure. Littlebox must do the conceptual work, not decorate the margin.

Read References As Needed

• references/visual-language.md: style, palette, density, and hard bans.
• references/littlebox-ip.md: closed-only Littlebox form, action vocabulary, and failure modes.
• references/composition-methods.md: composition families and metaphor invention rules.
• references/language-and-labels.md: multilingual output and in-image label handling.
• references/prompt-template.md: generation and editing prompt templates.
• references/quality-gate.md: checks before delivery or regeneration.
• references/examples.md: realistic invocation examples and forward-test cases.

Workflow

1. Understand the Source

Read the user's article, notes, screenshot, file, link, or pasted idea. Identify:

• the single strongest insight
• any cognitive turn, contrast, bottleneck, loop, handoff, or failure point
• which parts deserve images and which should stay text-only
• the likely language for visible labels

Do not illustrate every paragraph. Prefer high-leverage moments: decision points, before/after states, intake-to-output systems, evidence accumulation, scattered-to-structured transformations, hidden bottlenecks, recurring mistakes, and final handoff. Simplify only after the idea is fully expressed: clarity and semantic completeness outrank a fixed object count or a large empty area.

2. Plan Before Generating When Asked

If the user asks for planning, analysis, "where should I add images", or a shot list, return a concise shot list before creating images. For each proposed image include:

• placement in the source
• core idea
• Littlebox state: always closed
• visual metaphor
• background: pure white #FFFFFF by default, or the user's exact requested color
• suggested objects
• composition structure and visual counterweight
• short visible labels

Default to 4-7 images for a normal article, 1-3 for short text, and no more than 9 unless the user explicitly wants a full visual series.

When the user does not mention a background, assign pure white #FFFFFF to every image, including multi-image sets. If the user asks for alternatives, offer very pale cream #FFFCF5, pale sky blue #E3F2FD, and pale lavender #E6E6FA. If the user specifies any other background color, use that exact color instead. Do not randomize or balance background colors unless the user explicitly asks for variation.

3. Generate When Asked

If the user asks to generate, draw, make, or output images, use the available image generation tool. Generate each image separately; never pack multiple article illustrations into one image unless the user asks for a comparison board.

For each image:

• use one core idea only, but include every object needed to make that idea understandable
• make Littlebox perform the central action
• keep Littlebox closed in every final image; express intake, sorting, filtering, and transformation with external slips, stacks, paths, stamps, props, and side-seam twig arms
• keep Littlebox anatomically canonical: exactly two possible twig arms total, one from each side seam; never add a third arm to handle extra categories
• choose or preserve the assigned background from the planning step; default to pure white #FFFFFF, and let an explicit user-specified color override every preset
• usually use 3-7 supporting elements; use fewer or more when the concept genuinely needs it, provided the hierarchy stays legible
• use short visible labels in the user's language
• distribute visual weight across the canvas with staggered heights, depth, scale, and counterweights; avoid placing all elements on one horizontal track
• choose a non-linear composition by default: orbit, diagonal tension, triangular balance, staggered clusters, field composition, or asymmetric counterweight
• keep breathing room around meaningful groups, but never delete necessary content just to manufacture emptiness

Use references/prompt-template.md for the final prompt wording.

4. Review And Iterate

After generation, inspect the image against references/quality-gate.md. Regenerate or edit if:

• Littlebox is passive
• labels are too long or unreadable
• the result looks like a formal slide, glossy mascot, vector infographic, or logistics icon
• Littlebox is open, half-open, cutaway, or using raised flaps
• color is overused
• the image has no clear hierarchy or the supporting details compete with the core idea

5. Save Deliverables

When working inside a project workspace, save final images under:

assets/<article-slug>-littlebox/

Name files in order:

01-topic-name.png
02-topic-name.png

Keep generated source files unless the user asks to delete them. Report saved paths and identify which images are final, optional, or need one more iteration.

Output Style

Keep responses compact. Use the user's language for planning and delivery notes. Use English for image-generation prompts, with exact visible labels quoted in the user's language.