Scientific illustration, from paper to figure — an all-in-one prompt engine. sci-plot does not generate images directly; it produces cross-platform prompts that you copy into ChatGPT/DALL-E, Doubao, Qwen (Tongyi Wanxiang), Midjourney, Stable Diffusion, or any other AI image generation tool.
Works with any AI coding assistant — Claude Code, Codex, Cursor, Windsurf, GitHub Copilot, Cline, Continue, Aider, or any AI that can read project files. Just open this directory and the AI will understand SKILL.md instructions.
Your paper is done — time for figures. Your LaTeX paper (.tex file) is complete: methods, formulas, experiments are all written, but you lack a high-quality method architecture diagram or mechanism flowchart. Point sci-plot at your project directory or .tex file. It reads everything, proposes a figure plan, you confirm the parameters, and it outputs prompts for each figure. Copy the prompts to ChatGPT/DALL-E / Doubao / Qwen / Midjourney — images appear in seconds.
Best results come from complete content. The more complete and well-structured your input is, the more precise the extraction and the better the prompts will match your paper's academic context. Vague one-line ideas still work, but the output will be more generic.
Also suitable for: project README illustrations, technical blog visuals, defense presentation slides, algorithm explanation diagrams.
git clone https://github.com/LPK3215/sci-plot.gitAfter cloning, open the sci-plot directory (or add it as a workspace/subproject). The AI will read SKILL.md and apply its rules.
See a complete example first: Check out examples/transformer-paper/ to see what sci-plot outputs look like — analysis, figure plan, and ready-to-use prompts for the classic Transformer paper.
After cloning, open the sci-plot directory and send this prompt to the AI:
Follow the /sci-plot/SKILL.md skill to generate scientific figure prompts.
Target: [optional — leave blank to auto-analyze current project]
Target supports any of the following formats:
- Leave blank → auto-analyze current project, prioritizing
.texfiles - Local path →
/path/to/your/paper/ - arxiv link →
https://arxiv.org/abs/9999.99999 - Pasted content → paste abstract, method description, or algorithm steps directly
- arxiv ID →
1706.03762
- AI analyzes your content and proposes a figure plan
- You confirm: language (Chinese/English/Bilingual), style, number of figures, aspect ratio, purpose
- AI generates prompt files into
sci-plot-output/prompts/directory - You copy prompts into ChatGPT/DALL-E, Doubao, Qwen, Midjourney, Stable Diffusion, or other AI image tools
| Style | Effect | Best for |
|---|---|---|
| paper-figure (default) | Paper figure style, white background, vector modules, restrained colors | Journal figures, README hero images, technical displays |
| sketchnote | Warm research notes style, cream background, hand-drawn lines | Knowledge sharing, learning notes |
| journal-minimal | Nature/IEEE style, extremely minimal academic | Paper submission, defense |
| warm-notes | Bright handwritten style, approachable yet professional | Science communication, courses |
| business-research | Business research brief style, structured, data-driven | Industry analysis, investor presentations |
| Tool | Action |
|---|---|
| ChatGPT (recommended) | Paste into DALL-E |
| Doubao | Text-to-image → paste prompt |
| Qwen (Tongyi Wanxiang) | Text-to-image → paste prompt (English works best) |
| Midjourney | /imagine then paste prompt |
| Stable Diffusion | txt2img paste |
| Other image gen tools | Works equally well |
sci-plot-output/
├── analysis.md # Content analysis summary
├── figure-plan.md # Figure plan (what each page covers)
├── prompts/
│ ├── 01-overview.md # Page 1 prompt
│ ├── 02-mechanism.md # Page 2 prompt
│ └── ...
└── README.md # This README
sci-plot/
├── SKILL.md # AI instruction: 6-round mandatory workflow
├── README.md # User-facing documentation (English)
├── README_CN.md # User-facing documentation (Chinese)
├── LICENSE # MIT License
├── CHANGELOG.md # Version history (English)
├── CHANGELOG_CN.md # Version history (Chinese)
├── CONTRIBUTING.md # Contribution guidelines (English)
├── CONTRIBUTING_CN.md # Contribution guidelines (Chinese)
├── SECURITY.md # Security policy
├── .gitignore # Version control exclusions
├── .gitattributes # Line ending normalization
├── index.html # GitHub Pages entry (redirects to overview)
├── project_overview.html # Project overview dashboard entry
├── project_overview/
│ ├── styles.css # Dark theme styles
│ └── main.js # Interactive logic
├── examples/
│ └── transformer-paper/ # Worked example: analysis, plan & ready-to-use prompts
├── docs/
│ ├── architecture.svg # Workflow architecture diagram (English)
│ ├── architecture_cn.svg # Workflow architecture diagram (Chinese)
│ └── scripts/
│ └── generate_architecture.py # Architecture diagram generation script
└── references/
├── prompt-rules.md # Core prompt generation rules
├── analysis-guide.md # Content analysis guide
├── layout-rules.md # Page composition & role definitions
└── styles/
├── paper-figure.md # Paper figure style
├── sketchnote.md # Warm sketchnote style
├── journal-minimal.md # Nature/IEEE minimal style
├── warm-notes.md # Bright handwritten style
└── business-research.md # Business research brief style
Q: Can sci-plot generate images directly? A: No. sci-plot only produces prompts. You copy the prompts into ChatGPT/DALL-E, Doubao, Qwen, Midjourney, or other image generation tools. This approach makes prompts cross-platform compatible, not tied to any specific tool.
Q: Can I use it without a complete paper or project? A: Yes, but results will be less precise. sci-plot works best with complete content — the more complete and well-structured your input, the more accurate the extraction. A vague one-liner still produces prompts, but they'll be more generic.
Q: What input formats are supported?
A: LaTeX papers (.tex, highest priority), Markdown documents, PDF, Python source code, arxiv links or IDs, and pasted text content.
Q: Which part of the prompt file is the core output?
A: The ▶ English Prompt (core, copy directly) section in each prompt file is the core output. Copy it directly into any image generation tool. The Chinese explanations, composition descriptions, and content elements are for human reference.
Q: How do I add a new visual style?
A: Create a new .md file under references/styles/, follow the format of existing style files, then register it in the style tables in both SKILL.md and README.md. See CONTRIBUTING.md for details.
Q: What's the maximum number of figures? A: Minimum 1, maximum 10. The exact number depends on content complexity: simple content 1-3, medium 4-6, complex 6-10.
- LPK3215 — GitHub
MIT — see LICENSE for details.