@Fred834567: Anyone who does research knows that making figures is even more painful than running code. Once the data is done, getting stuck on colors, fonts, layout, and architecture diagrams can make you pull your hair out. Today I found four awesome tools for scientific figure creation — worth bookmarking right away. First one, Scientific Visualization. This thing is basically a journal quality inspector incarnate...
Summary
This article introduces four scientific figure tools: Scientific Visualization, the Visualization Agent in Academic Research Skills, Nature Figure, and Deepshare's CCF-Figure — solving pain points such as color and layout, selecting figures and generating code, stitching subplots, and drawing paper architecture diagrams.
View Cached Full Text
Cached at: 11/21/26, 12:10 AM
Everyone in research knows this: making figures is more painful than running code.
Data’s done, but you’re stuck on colors, fonts, layout, and architecture diagrams — enough to pull out all your hair.
Today I found four awesome tools for scientific figure-making. Straight into your bookmarks they go.
1. Scientific Visualization
This thing is basically a journal quality inspector incarnate. I used to draw bar charts and get slammed by reviewers because it wasn’t clear whether the error bars showed SEM or SD.
It comes with top-journal standards built right in, and supports color-blind-friendly and black-and-white print modes.
The best part: the exported PDF, EPS, and TIFF files meet resolution requirements out of the box, saving you the agony of endless rework.
https://github.com/Panda-dda/Scientific-Visualization…
2. The Visualization Agent in Academic Research Skills
Not sure what kind of chart fits your data? It picks the chart based on your data’s characteristics and generates runnable Python or R code directly.
It even writes the LaTeX figure-insertion code for you — plotting and typesetting in one step.
https://github.com/Imbad0202/academic-research-skills…
3. Nature Figure
Designed specifically for the problem where a bunch of subplots stitched together look like a hodgepodge. It first asks you what your core conclusion is, then gives each subplot its own job to do.
Once generated, it automatically checks for text occlusion and margin alignment — if things don’t line up, it won’t let you export.
The whole point is a clear chain of evidence that a non-specialist can grasp at a glance.
https://github.com/Yuan1z0825/nature-skills/tree/main/skills/nature-figure…
4. CCF-Figure, open-sourced by 深度之眼 (DeepShare)
Architecture diagrams for AI papers are the biggest headache. It first determines which of seven paper categories yours falls into, then matches a layout from eleven structural templates.
Architectural loops and feedback classifications are all covered, and it finally outputs bilingual (Chinese/English) prompts you can hand straight to a large model to render the figure — plus a self-check checklist for common failures.
https://github.com/Deepshare-Official/CCF-Figure…
Panda-dda/Scientific-Visualization
Source: https://github.com/Panda-dda/Scientific-Visualization
Scientific Visualization
Scientific plotting workflows for AI coding agents.
English | 简体中文
Create, refine, and review scientific figures with reproducible Python/Matplotlib code. Describe the data and the intended result; the skill guides an agent through method selection, implementation, visual inspection, and export.
11 reference modules · 25 techniques · 4 runnable examples · PDF / SVG / PNG / PGF / TEX
Quick start · Usage · Examples · Validation · Licensing
Gallery
All previews below were generated by the included examples using explicitly labeled synthetic data.
| Line plot | Scatter plot with marginal distributions |
|---|---|
| Synthetic damped signal with labeled time and signal axes | Synthetic paired samples with aligned count histograms |
| Orthographic mesh projection | Dense scatter with selective rasterization |
|---|---|
| Triangulated cube projected onto equal-scale axes | Dense synthetic scatter with separate text and axis layers |
Capabilities
- Data-aware plotting. Check shapes, units, missing values, and axis domains before choosing a visual encoding.
- Figure refinement. Improve typography, colors, legends, annotations, and spacing while preserving data meaning.
- Multi-panel layouts. Organize shared axes, marginal distributions, and panels with different proportions.
- 3D geometry. Render triangular meshes with orthographic projection and depth ordering.
- Publication export. Specify physical dimensions and export PDF, SVG, PNG, PGF, and TEX; rasterize dense data layers when appropriate.
- Traceable execution. Record data provenance, parameters, versions, export results, and the actual inspection status.
The skill uses progressive disclosure: SKILL.md provides the workflow and routes each task to relevant references. Detailed knowledge stays outside the entry point.
The current executable implementation targets Python/Matplotlib. Codex discovery and execution after installation have been checked. The skill instructions and technical references are primarily in Chinese; this repository provides English and Chinese project READMEs.
Quick start
1. Clone the repository
git clone https://github.com/Panda-dda/Scientific-Visualization.git
cd Scientific-Visualization
2. Install the skill in Codex
Copy the repository contents into a directory named scientific-visualization under ~/.agents/skills/.
Run this command from the repository root. It uses Python for cross-platform copying and refuses to overwrite an existing installation:
python -c "from pathlib import Path; import shutil; shutil.copytree(Path.cwd(), Path.home()/'.agents'/'skills'/'scientific-visualization', ignore=shutil.ignore_patterns('.git', '.venv', '__pycache__', 'outputs', '.env', '.env.*'))"
The installed layout should include ~/.agents/skills/scientific-visualization/SKILL.md directly. For project-only use, copy it into that project’s .agents/skills/scientific-visualization/ instead. Choose one scope to avoid duplicate entries.
Invoke the skill with $scientific-visualization. Reload or restart Codex if the new skill does not appear. See the official skill documentation (https://learn.chatgpt.com/docs/build-skills) for discovery behavior.
3. Prepare a plotting environment
The core examples require Python, NumPy, and Matplotlib. PDF/SVG/PNG run without the source book repository, network access, external fonts, or LaTeX.
Use an existing environment, or create one:
python -m venv .venv
Activate it with .venv\Scripts\Activate.ps1 in Windows PowerShell, or source .venv/bin/activate on macOS/Linux, then install the plotting dependencies:
python -m pip install numpy matplotlib
This installs available package versions; the versions actually tested are listed under Validation.
4. Describe the figure
$scientific-visualization
Read data.csv. Column t is time in seconds; x1 through x5 are displacements in meters.
Plot all five series with consistent colors and a legend that does not obscure the data.
Use a 120 mm × 85 mm canvas and export PDF, SVG, and PNG into figures/.
Execute the code, inspect the generated figure, and correct any layout problems.
For a first trial without a dataset:
$scientific-visualization
Generate a line plot, a scatter plot with marginals, and a projected 3D mesh.
Use explicitly labeled synthetic data, save the outputs to demo-figures/,
and inspect the resulting figures.
Usage
Choose output formats explicitly, for example --formats tex or --formats pdf png. Choices: PDF, SVG, PNG, PGF, TEX (case-insensitive); omitted formats default to PDF/SVG/PNG. TEX generates a standalone .tex wrapper and required .pgf drawing, plus PNG sidecars for raster layers. TEX/PGF require XeLaTeX with TikZ/PGF; compiling the wrapper also requires the standalone package. Keep the bundle together and run xelatex figure.tex from its directory. This is PGF drawing code, not editable PGFPlots data. No PDF preview is generated unless requested separately.
Provide the data or plotting script, the meaning and units of each variable, and the intended output. If the chart type is uncertain, ask the agent to inspect the data and propose an appropriate representation.
Choose TEX output
$scientific-visualization
Plot my data and export only TEX, including any required companion files.
python examples/line_scatter.py --demo --formats tex --output-dir outputs/tex
cd outputs/tex
xelatex -interaction=nonstopmode -halt-on-error line_scatter.tex
The export step writes line_scatter.tex, line_scatter.pgf, and run.json. The separate compilation step creates a PDF and LaTeX auxiliary files.
Refine existing code
$scientific-visualization
Review plot_results.py and its current output. Improve typography, colors,
legend placement, and panel spacing. Preserve the data processing and physical
meaning. Save new outputs and explain the changes.
Build a multi-panel figure
$scientific-visualization
Use results.csv for a two-column figure: state trajectories on the left and
the precomputed error quantity on the right. Confirm definitions and units,
align the time ranges, and export a 180 mm-wide PDF and PNG.
Review without editing
$scientific-visualization
Review this figure for readability and faithful data representation.
Identify specific problems with scales, units, legends, and color encoding.
Do not modify the code or data yet.
Keep an existing toolchain
For a MATLAB or R task, explicitly ask the agent to preserve that language. General design principles can transfer, but this repository does not provide or validate equivalent implementations for those toolchains.
Run the examples
The examples also work independently of an agent. From the repository root:
python examples/line_scatter.py --demo --output-dir outputs/line
python examples/marginal_panels.py --demo --seed 7 --output-dir outputs/marginals
python examples/projected_mesh.py --demo --output-dir outputs/mesh
python examples/mixed_export.py --demo --output-dir outputs/export
Use a CSV with explicit column names and units:
python examples/line_scatter.py --input data.csv --x-column t --y-column value --xlabel "Time (s)" --ylabel "Displacement (m)" --output-dir outputs/my-data
| Option | Purpose |
|---|---|
--input / --demo | Choose a file or synthetic demonstration; mutually exclusive |
--synthetic-input | Label an external synthetic dataset in the figure and log |
--output-dir | Set a dedicated output directory |
--formats tex | Select one or more of pdf svg png pgf tex; default: pdf svg png |
--width-mm / --height-mm | Set physical canvas dimensions |
--dpi | Set raster resolution, including raster layers inside vector files |
--font-size / --line-width | Override style defaults |
--overwrite | Explicitly permit replacing target outputs |
--help | Show all options for an example |
CSV values must be finite numbers. Invalid cells are reported with row and column information. NPZ mesh input requires vertices: (N, 3) and zero-based integer faces: (M, 3); --unit identifies the coordinate unit.
Each run writes figures and a run.json record containing provenance, input hash, parameters, dependency versions, reference IDs, and export status. Visual inspection starts as not_performed and must be updated only after an actual review. Missing inputs never silently fall back to synthetic data. Existing outputs are protected unless --overwrite is supplied.
Repository layout
Scientific-Visualization/
├── SKILL.md # Entry point, workflow, and reference routing
├── references/ # 11 focused reference modules
├── examples/ # 4 examples and shared input/export utilities
├── assets/styles/ # Configurable Matplotlib style
├── docs/
│ ├── images/ # Synthetic example previews
│ └── validation.md # Validation summary (Chinese)
├── SOURCE_MAP.md # Technique IDs, source locations, adaptations
├── README.zh-CN.md # Chinese project overview
├── LICENSE.md # File-level licensing scope
└── LICENSES/ # MIT, upstream terms, and attribution notices
Validation
Results recorded on 2026-11-06:
| Area | Observed result |
|---|---|
| Environment | Windows, Python 3.12.7, NumPy 1.26.4, Matplotlib 3.9.2 |
| Skill integration | Format validation, Codex CLI 0.154.0 discovery, execution after installation |
| Behavioral checks | 44 checks passed across input validation, scale domains, projection, export, and overwrite protection |
| Portability within the tested environment | Four examples ran from an isolated copy and a different working directory |
| PDF output | Six PDFs rendered with Poppler and visually reviewed; physical dimensions and font embedding checked |
| TEX / PGF | Line and rasterized scatter bundles exported and compiled with XeLaTeX (TeX Live 2025); visual inspection pending |
| PNG / SVG | PNG dimensions checked; SVG XML, text, and mixed-layer structure checked |
See the validation notes for scope and limitations. These are observations from a specific environment, not a compatibility guarantee for every platform or dependency version.
Known limits
- Painter’s depth ordering does not resolve every intersecting or transparent mesh.
- Synthetic consensus and formation examples validate plotting adaptations, not control-system performance.
- Statistical inference, error definitions, and stability claims require task-specific evidence.
- Chinese font fallback, custom LaTeX fonts/preambles, Lab-color recipes, and MP4 encoding have not completed dedicated acceptance testing.
- Implicit skill selection remains enabled by default, but independent end-to-end selection testing is not complete.
- Style defaults are adjustable starting points; journal-specific requirements take precedence.
Contributing
Issues with a minimal dataset or plotting script are welcome. Include the command, Python/Matplotlib versions, expected behavior, and actual output.
When adding a method or example:
- Define its inputs, units, and applicability.
- Preserve attribution and update
SOURCE_MAP.md. - Label synthetic data explicitly.
- Report execution and visual inspection separately.
- Recheck affected examples when shared styles or export logic change.
Licensing and attribution
This repository uses component-specific licenses. The complete scope is defined in LICENSE.md.
| Material | License |
|---|---|
Original project READMEs, validation notes, licensing overview, and .gitignore | MIT |
| Book-derived instructions, references, source mapping, and style | CC BY-NC-SA 4.0, with source-specific exceptions retained |
| Adapted example code | Applicable source terms, including BSD 2-Clause |
| Generated example previews | CC BY-NC-SA 4.0 |
The MIT grant applies only to the explicitly listed original project materials. It does not relicense the book, reference snippets, example code, or embedded preview images.
Methods are informed by Nicolas P. Rougier’s Scientific Visualization: Python + Matplotlib (https://github.com/rougier/scientific-visualization-book). This project organizes selected techniques into task-oriented references, agent workflows, and parameterized examples, with additional input validation, output protection, and execution records.
Source baseline: 62fa569f30333c817c13e4dc757877c1192fd15a. See SOURCE_MAP.md and NOTICE for provenance and adaptations. This is an independent project and does not imply endorsement by the original author.
The repository does not redistribute the book PDF, font packages, original model assets, or the book’s pre-rendered figures.
Similar Articles
@VincentLogic: Discovered an amazing tool for scientific visualization! nature-skills lets you create Nature-level figures using matplotlib. It handles multi-panel layouts, color schemes, fonts, and typography seamlessly, directly exporting editable SVGs. Comes with 5 built-in Nature-style template examples, supporting bar…
nature-skills is a scientific visualization tool based on matplotlib. It features built-in Nature-style templates, supports various chart types, and outputs editable SVG format, aiming to streamline the figure creation process for scientific papers.
@QingQ77: Automatically convert charts from scientific papers into executable Python/matplotlib plotting code. https://github.com/littlepeachs/NaturePanelForge… A tool to convert scientific paper charts into code. Give it a Natur…
This is a tool that automatically converts scientific paper charts into executable Python plotting code, using the Qwen vision model and Codex agent for panel segmentation, code generation, and refinement.
@GitHub_Daily: Need to explain project system architecture to colleagues. Just talking without diagrams is limited, and drawing manually is time-consuming and often not well-made. archify, an Agent Skill that can be embedded into Claude Code, Codex CLI, and opencode, turns plain language descriptions directly into an architecture diagram…
archify is an Agent Skill that can be embedded into Claude Code, Codex CLI, and opencode, directly generating system architecture diagrams, workflow diagrams, sequence diagrams, data flow diagrams, and lifecycle state diagrams from plain language descriptions. It supports dark/light theme toggling and export in multiple formats.
@grgerwcwetwet: Found an awesome open-source Skill! No more messing around with Figma for flowcharts and architecture diagrams! Recommending a GitHub project: Diagram Design — let Claude Code, Codex, and Cursor generate professional-grade diagrams for you. 1. Built-in 4…
Recommending an open-source Skill project called Diagram Design, which lets Claude Code, Codex, and Cursor automatically generate professional diagrams such as flowcharts and architecture diagrams. It includes 45 diagram types, supports brand color matching, and exports to HTML/SVG.
@vintcessun: Finally, someone has paved the way for AI-assisted research, connecting the once fragmented pipeline. Writing papers, compiling literature reports, generating simulation figures, and creating defense slides — in the past each step required manually shuffling data and copying results. Now, three Skills tie the entire workflow together: from scientific-toolkit for computing data and generating figures, feeding directly into research-writing for the main text, and then letting office-academic produce editable PPTs and Word documents, completely eliminating repetitive manual effort. The core is a complementary collaboration set, installable with one click in Claude Code and Codex, with Chinese-first support.
A report on an AI-assisted research tool that connects the entire research workflow through three Skills (scientific-toolkit, research-writing, office-academic), from data computation to paper writing to PPT creation. It supports one-click installation in Claude Code and Codex, with Chinese-first priority.