Generate for academic papersextremely detailed English prompts, for AI image generation tools (NanoBanana / Gemini / Midjourney / DALL-E) to generate top-conference-level professional academic figures.
Core Philosophy
The generated prompts must achieve three things:Extremely high information density、Precise visual style、Complete content with no omissions。
It is better for the prompt to be too long and too detailed than to simplify or omit anything. The value of an academic figure lies in precisely conveying complex information, not in being aesthetically pleasing and concise.
Workflow
Step 1: Understand the paper content
Before generating prompts, youmustfirst fully understand the paper content:
- Read the paper/chapter source files provided by the user (LaTeX, Word, PDF, etc.)
- Extract the core concepts, methods, model architecture, and data flow of each section
- Identify all locations that need figures and their content requirements
- Understand the mathematical notation, variable meanings, and dimension information in the paper
Step 2: Analyze reference figures (if any)
If the user provides reference figures, youmustanalyze in detail:
| Analysis Dimension | Extracted Content |
|---|---|
| Color scheme | Precise color values for primary, secondary, and accent colors (e.g., #3A8F85) |
| Layout structure | Flow direction (left→right / top→bottom), partitioning method, hierarchical relationships |
| Module style | Box shape, border thickness, fill color, corner radius |
| Annotation method | Title bar style, formula annotations, dimension annotations, arrow types |
| Information density | Number of sub-details within each module, usage of embedded thumbnails |
| Special elements | Feedback loops, dashed boxes, jump arrows, legend position |
Step 2.5: Color scheme selection (must be completed before generating prompts)
Before generating any prompt, you must first show the user the color scheme options and wait for confirmation.
Show the following:
Please select a color scheme (enter a number or customize):
| # | Scheme Name | Style Positioning | Primary Color | Secondary Color | Accent Color |
|---|---|---|---|---|---|
| A | Okabe-Ito Academic Standard | Recommended by Nature / Science / CVPR, colorblind-friendly | Steel Blue #0072B2 |
Warm Orange #E69F00 |
Bluish Green #009E73 |
| B | Blue monochrome | Restrained, suitable for detailed module diagrams | Navy #0072B2 |
Medium Blue #4A90D9 |
Light Blue #A0C4E8 |
| C | Teal + Amber | Strong modern feel, ICLR / NeurIPS style | Deep Teal #00897B |
Amber #FFB300 |
Soft Grey #ECEFF1 |
| D | Navy + Coral | Steady and grand, IEEE journal style | Deep Navy #1A3A5C |
Coral #E05A47 |
Warm Sand #F5ECD7 |
| E | Slate + Violet | Elegant cool tones, medical / bioinformatics style | Slate Blue #3F51B5 |
Muted Violet #7E57C2 |
Pale Lavender #EDE7F6 |
| F | Forest + Gold | Heavy academic feel, natural science journal style | Forest Green #2E7D32 |
Gold #C49A00 |
Cream #F9F6EE |
| G | Minimal Grey | Minimal grayscale + single accent color, arXiv technical report style | Charcoal #263238 |
Steel #546E7A |
Single accent (user-specified) |
| H | Custom | Color values provided by the user or selected from the tools below | — | — | — |
For custom color schemes, the following tools are recommended:
- Coolors — Random generation + lock adjustment, export palettes:https://coolors.co
- ColorHunt — Curated high-quality palettes, supports tag filtering:https://colorhunt.co
- Adobe Color — Color wheel + complementary/analogous/triadic color scheme generation:https://color.adobe.com/create
- ColorBrewer — Designed specifically for academic data visualization, supports color blindness safety verification:https://colorbrewer2.org
- Viz Palette — Designed specifically for data visualization color schemes, real-time simulation of color blindness effects:https://projects.susielu.com/viz-palette
- Paletton — Color wheel-driven color scheme designer:https://paletton.com
Tip: After choosing colors, simply tell me the hex values of the primary/secondary/accent colors (e.g.,
主色 #2E7D32，辅色 #C49A00）。
After waiting for the user to choose, proceed to Step 3 to generate prompts.
If the user has already explicitly specified the color scheme in the initial request (e.g., "use a blue-green color scheme", "refer to my reference image"), skip this step and proceed directly to Step 3.
Step 3: Generate Prompts
Generate a prompt for each image according to the "Prompt Structure Template" below.
Prompt Structure Template
Each prompt must contain the following four levels, none may be omitted:
Level 1: Global Description
An opening paragraph summarizing the type, theme, and overall layout of the entire image.
A highly detailed, information-dense academic paper [类型] diagram in the style of
top-tier [目标会议] publications. The diagram illustrates [主题概述], arranged as
[布局描述: e.g., "a rich multi-stage left-to-right pipeline with multiple parallel
pathways, embedded thumbnail visualizations, and dense annotations"].
Type Vocabulary:
- architecture / framework / pipeline / flowchart / comparison / ablation
- network architecture / module detail / data flow / system overview
Level 2: Section-by-Section Description
Use the === SECTION NAME === format to separate each region. Each region must internally contain:
| Element | Requirement | Example |
|---|---|---|
| Background panel | Very light color fill + color value | very faint grey #F7F7F7 background panel |
| Section label | small-caps text + thin gray divider line | small-caps Steel Blue label "ENCODER" |
| Module box | White fill + colored/gray border | white rounded-rectangle box with Steel Blue border |
| Substructure | Components inside each module | containing three parallel branches... |
| Embedded thumbnail | Small monochrome visualization inside the module | a small monochrome FFT spectrum bar chart thumbnail |
| Formula annotation | LaTeX-style formula | with formula "HT = 1/(1 + d/d₀)" |
| Dimension annotation | Tensor/vector dimensions | labeled "X ∈ R^(120×6)" |
| Connecting arrow | Arrow type and label | thin arrow labeled "30-step predicted trajectory" |
Key principle: Every module box must have sub-content inside. Blank placeholder boxes are not allowed.
Level 3: Global Annotations
- Dimension annotations along the main arrows:
"R^(120×6)", "R^(30×6)", "R^14", "R^5" - Feedback loops (if any):
dashed feedback arrow from output back to input - Legend (if any):
legend box in bottom-right corner - Cross-region connections:
skip connection dashed arrow from Stage 1 to Stage 3
Level 4: Style Specifications
A complete style description must be appended at the end of each prompt, filling in the color values of the selected scheme.
Color Scheme Color Value Table
After the user selects a scheme in Step 2.5, take the corresponding color values from below and fill them into the STYLE SPECIFICATIONS at the end of the prompt.
Scheme A: Okabe-Ito Academic Standard (Default)
| Element | Color value | Usage |
|---|---|---|
| Primary color | Steel Blue #0072B2 |
Core module borders, section labels, main arrows |
| Secondary color | Warm Orange #E69F00 |
Secondary module borders, alternative highlights |
| Accent color | Bluish Green #009E73 |
Output/result modules (very少量) |
| Warning color | Vermillion #D55E00 |
Key highlights/warnings (very少量) |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Region background | Faint Grey #F7F7F7 |
Large region grouping |
| Standard border | Light Grey #CCCCCC |
Normal frame (1px) |
| Body text color | Charcoal #333333 |
All labels |
| Arrows/Lines | Dark Grey #4D4D4D |
Connector lines |
| Secondary annotations | Medium Grey #666666 |
Annotation text |
Scheme B: Blue monochrome
| Element | Color value | Usage |
|---|---|---|
| Accent/Label | Navy #0072B2 |
Key module borders, section labels |
| Primary border | Medium Blue #4A90D9 |
Main flow modules |
| Secondary border | Light Blue #A0C4E8 |
Secondary/auxiliary modules |
| Area background | Pale Blue #EBF3FA |
Large area grouping |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Neutral/Arrow | Grey #999999 |
Connector lines, secondary annotations |
| Body text color | Charcoal #333333 |
All labels |
Scheme C: Teal + Amber
| Element | Color value | Usage |
|---|---|---|
| Primary color | Deep Teal #00897B |
Core module borders, section labels |
| Secondary color | Amber #FFB300 |
Secondary module borders, emphasis |
| Area background | Soft Grey-Blue #ECEFF1 |
Large area grouping |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Standard border | Light Grey #CFD8DC |
Normal frame |
| Body text color | Dark Slate #263238 |
All labels |
| Arrows/Lines | Medium Grey #607D8B |
Connector lines |
| Secondary annotations | Grey #90A4AE |
Annotation text |
Scheme D: Navy + Coral
| Element | Color value | Usage |
|---|---|---|
| Primary color | Deep Navy #1A3A5C |
Core module borders, section labels |
| Secondary color | Coral #E05A47 |
Emphasis modules, critical paths |
| Accent color | Warm Sand #D4A96A |
Output/result modules (very few) |
| Area background | Ivory #F9F6EE |
Large area grouping |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Standard border | Light Warm Grey #D4C5B0 |
Normal frame |
| Body text color | Dark Slate #1A3A5C |
All labels |
| Arrows/Lines | Medium Grey #7A8A99 |
Connector lines |
Scheme E: Slate + Violet
| Element | Color value | Usage |
|---|---|---|
| Primary color | Indigo #3F51B5 |
Core module borders, section labels |
| Secondary color | Muted Violet #7E57C2 |
Secondary module borders |
| Accent color | Teal Accent #26A69A |
Output/result modules (very few) |
| Area background | Pale Lavender #EDE7F6 |
Large area grouping |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Standard border | Light Purple Grey #C5CAE9 |
Normal frame |
| Body text color | Deep Indigo #1A237E |
All labels |
| Arrows/Lines | Medium Grey #7986CB |
Connector lines |
Scheme F: Forest + Gold
| Element | Color value | Usage |
|---|---|---|
| Primary color | Forest Green #2E7D32 |
Core module borders, section labels |
| Secondary color | Antique Gold #C49A00 |
Secondary module borders, accents |
| Area background | Warm Cream #F9F6EE |
Large area grouping |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Standard borders | Sage #A5D6A7 |
Regular frames |
| Body text color | Dark Forest #1B5E20 |
All labels |
| Arrows/lines | Warm Grey #795548 |
Connecting lines |
Scheme G: Minimal Grey + single accent color
| Element | Color value | Usage |
|---|---|---|
| Accent color (replaceable) | Steel Blue #0072B2 |
Only used for the borders of the 1-2 most core modules |
| Dark grey | Charcoal #263238 |
Section labels, body text color |
| Medium-dark grey | Slate #546E7A |
Main borders, arrows |
| Medium grey | Medium Grey #90A4AE |
Secondary borders |
| Light grey | Light Grey #ECEFF1 |
Area background |
| Module fill | Pure White #FFFFFF |
All content boxes |
| Standard borders | Grey #B0BEC5 |
Regular frames |
Scheme H: User-defined
Extract from a reference image, or use a color picker tool to select color values, then provide them in the following format:
主色：#XXXXXX（核心模块边框/节标签）
辅色：#XXXXXX（次要模块/强调）
点缀色：#XXXXXX（输出结果，可选）
背景：#XXXXXX（区域分组背景，建议极浅）
文字：#XXXXXX（建议深色）
Color taboos (avoid an AI-generated look)
| Prohibited practice | Alternative practice |
|---|---|
| 4-5 colored background panels | Predominantly white + very light grey grouping |
| High-saturation Header Banner Bar | Small small-caps text labels + grey divider lines |
| Different color fill for each module | Pure white fill + color only for borders |
| Colored thumbnails | Monochrome grayscale or only 2 colors |
| 5+ colors appearing simultaneously | At most 3 colors + grey tones |
| Rainbow/gradient effects | Solid color, flat, no gradient |
Image type-specific templates
Type 1: Overall Framework
结构: [输入] → [阶段1] → [阶段2] → ... → [输出]
要求:
- 各阶段用 small-caps 文字标签 + 细灰色分割线（不用彩色背景面板）
- 核心模块用主色边框，次要模块用辅色边框，普通模块用 Grey 边框
- 所有模块白色填充，无彩色填充
- 阶段间箭头标注数据维度和含义
- 每个阶段内展示 2-4 个子模块，子模块内嵌入单色缩略图
- 底部或顶部添加反馈/跳接连接（灰色虚线）
Type 2: Network Architecture
结构: [输入层] → [编码器堆叠] → [核心模块（并行分支）] → [输出头]
要求:
- 并行分支用不同边框色区分，内部均为白色填充
- 每个分支内标注操作名称 + 单色缩略图
- 残差连接用灰色虚线弧形箭头
- 维度标注在每层转换处（灰色小字）
- 重复层用 "×N" 灰色虚线框标识
Type 3: Module Detail
结构: [输入] → [操作1] → [中间表示] → [操作2] → [输出]
要求:
- 每步操作用白色独立框，关键操作用主色边框，次要用 Grey 边框
- 中间表示用单色灰度缩略图可视化（频谱图、热力图等）
- 跳接/残差连接用灰色虚线
- 公式标注在对应操作框旁（灰色小字）
Type 4: Comparison / Ablation
结构: N 列并排，每列一个变体
要求:
- 共享的基础结构用相同灰色边框
- 差异部分用主色或辅色边框高亮 + 灰色虚线框标出
- 每列顶部标注变体名称（small-caps）
- 底部可添加性能指标对比条（仅用主色 + 辅色 + Grey）
Type 5: Data / Behavior Patterns
结构: 1×N 网格，每格一个类别
要求:
- 每格为白色框 + 细灰色边框，顶部标签用主色小字
- 格内嵌入该类别的单色/双色典型可视化（轨迹、波形等）
- 用主色 / 辅色区分正面/负面类别（如有对比语义）
- 关键特征用文字标签
- 底部共享坐标轴（如有）
Thumbnail Vocabulary
| Data type | Thumbnail description |
|---|---|
| Time series | a small time-series waveform thumbnail |
| Frequency spectrum | a small frequency spectrum bar chart thumbnail |
| Attention map | a small monochrome attention heatmap grid thumbnail |
| Trajectory | a small 3D orbital trajectory curve thumbnail |
| Probability distribution | a small probability distribution bar chart thumbnail |
| Decision tree | a small decision tree branching diagram thumbnail |
| Confusion matrix | a small confusion matrix heatmap thumbnail |
| Network nodes | a small neural network layer diagram thumbnail |
| Feature vector | a small horizontal feature vector bar visualization |
| Scatter plot | a small scatter plot with cluster coloring thumbnail |
| Receptive field | a progressively larger receptive field grid icon |
| Convolution kernel | a small convolution filter kernel grid thumbnail |
| Gradient flow | a small gradient flow direction arrow diagram |
| Loss curve | a small training loss convergence curve thumbnail |
| ROC curve | a small ROC curve with AUC shading thumbnail |
| Image sample | a small example image/photo thumbnail |
| Point cloud | a small 3D point cloud visualization thumbnail |
| Heatmap | a small spatial heatmap overlay thumbnail |
Quality checklist
After generating each prompt, self-check against the following checklist:
- Information density: every module box contains sub-content (sub-boxes, thumbnails, formulas), with no blank boxes
- Restrained color: Use only 2-3 colors, with no extra colors
- White-dominant: ≥70% of the area is white/near-white, with no colored background panels
- Borders rather than fills: Modules use white fill + thin colored/gray borders, rather than colored fills
- Sectioning method: Use small-caps text labels + gray divider lines, not colored banner bars
- Dimension annotations: Dimensions are annotated on all major data flow arrows (e.g., R^(N×D))
- Formula annotations: Corresponding mathematical formulas are placed next to key operations
- Thumbnail embedding: At least 50% of modules have embedded monochrome/duotone thumbnail visualizations
- Completeness: All components described in the paper are represented in the figure, with no omissions
- Clear connections: Parallel paths, residual connections, and feedback loops are all clearly described
- Style specifications: Include a complete STYLE SPECIFICATIONS section at the end (including color value constraints and prohibited items)
- No simplification: Do not omit any content using "..." or "etc."
- Grayscale test: Ensure the description guarantees the image remains fully readable when printed in black and white
Output format
Wrap each prompt in a markdown code block:
### 图 X.Y — [中文图名]
适用类型：[框架图/架构图/模块图/对比图/模板图]
配色方案：[已选方案名]
推荐分辨率：[建议的宽高比，如 16:9, 3:2]
```
[完整英文提示词]
```
Notes
- Prompt language: The prompt itself must be in English, while explanatory text is in Chinese
- Unlimited length: Prefer longer over shorter; information density is the first priority
- Domain adaptation: Adjust thumbnail and icon choices according to the paper's domain (CV, NLP, Robotics, medicine, etc.)
- Reference images take priority: If the user provides reference images, the color scheme and layout should follow the reference images, overriding the preset scheme
- Batch generation: When the user requests figures for an entire paper, organize them by section and provide priority recommendations