Generate Guizang-style material illustrations, labeled explanatory visuals, material-styled chart illustrations, and data-first editorial images from articles, notes, product concepts, workplace reports, creator posts, tutorials, school materials, humanities topics, science explanations, screenshots, or chart data.
SkillScore77Maintained
Automatic
◆Design & UX Review pack
Last verified 8/14/2026
What it changes
Passed 16 of 21 test cases with the skill, and 11 of 21 without it.
Without skill55%
With skill77%
Checks it passed that the model alone missed
The prompt plan explicitly specifies the 3D Swiss editorial visual style
The prompt plan selects the Pipeline visual structure for the ordered multi-step workflow
The plan selects the Data-first scene structure to embed metric panels into a topical scene
and 4 more in the full report
6 fixed11 passed either way4 still failing1 broke
Pass-rate lift
+23 pp
Tokens
+26%
Agent turns
0%
Response time
-22%
Model gemini-3.6-flashMeasured 8/14/2026v117 of 22 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
I want diagram labels inside the image artwork explaining the user verification step. Should I use '用户在前端界面中输入手机验证码并点击确认提交按钮' as an in-image label text string?
What was checked, without and with the skill
Checked
Without skill
With skill
The response recommends against long explanatory sentences inside the image artwork
Passed
Passed
The suggested internal label text is constrained to short Chinese phrases between 2 and 5 characters
Missed
Passed
case-13 · gemini-3.6-flash · 8/14/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md82 lines8 files
Guizang Material Illustration
Create supporting illustrations from source text, screenshots, or chart data. The output is an image-generation prompt plan plus generated raster images when image generation is available.
This skill focuses on the illustration layer, not the full social-card layout. If the user also needs Xiaohongshu/WeChat card composition, pair this skill with the social-card skill: use this skill for the central illustrations, then place the generated images in the card template.
This skill does not replace the user's PPT skill or the Guizang social-card text-layout skill. Use those skills for slide structure and 3:4 text/card layout. Use this skill for the visual asset that goes into those layouts.
Workflow
Read the user's source text, screenshot, or data and identify the concepts or charts that deserve supporting images.
Decide the working mode yourself from context. Do not ask the user to choose a mode unless the missing choice would materially change the result. If clarification is needed, ask naturally in one short sentence and offer a recommended default.
If the source is a chart screenshot, extract only chart type, title, data, axis labels, axis range, tick labels, units, category order, and error bars. Do not carry over screenshot colors, typography, spacing, shadows, or background.
If the concept, entity, object, historical/cultural context, scientific mechanism, species, material, brand, model, or place is likely unfamiliar or visually specific, look up reference information and/or reference images before prompting. Use references to understand and extract stable visual cues, then translate those cues into the Guizang visual system.
Compress each non-chart concept into one plain-language explanation and 3-5 visible diagram labels when labels help. Some editorial or humanities illustrations may need fewer labels.
Data-first scene: chart or metric panel embedded in a topical scene.
Scientific mechanism: object, parts, forces, reactions, or biological process.
Text scene: a literary, historical, or everyday scene that anchors an abstract idea.
Write one image prompt per illustration. Make the prompt describe exact label text or chart data, aspect ratio, safe margins, references used, and shared Guizang visual style.
Generate the images with the imagegen skill or built-in image generation tool when available.
Inspect each image. If labels are wrong, chart data is wrong, reference cues are misleading, unreadable, or clipped, regenerate with stricter constraints.
Save prompts and final image paths in the task folder so the image set can be reproduced.
Text Rules
Use everyday Chinese. Avoid literal field labels like Trigger, Stop, Use unless the user explicitly wants bilingual UI.
Keep labels short: 2-5 Chinese characters is ideal; 6 characters is usually the upper limit.
Put explanatory sentences outside the illustration when the final artifact has a surrounding layout. Inside the illustration, only label objects and flows.
Prefer concrete labels over abstract nouns. Use 用户提示, AI 执行, 结果检查, 下一轮; not 输入阶段, 执行阶段, 验证阶段.
If the model struggles with Chinese text, shorten labels further and use distinct positions: top-left, top-center, top-right, bottom-center.
Do not add a dense legend inside the image. If a viewer needs a paragraph, it belongs in HTML/CSS, Markdown, or slide text.