academic-figures

Publication-ready scientific figures from one command — 22 chart types (bar, grouped bar, scatter, heatmap, forest plot, KM survival curve (Kaplan-Meier), ROC, violin, box, composite panels, PRISMA 2020 flow, funnel, Bland-Altman, PCA, venn 2-4 sets, clustered heatmap, dual-axis, Cox multi-variable regression forest), 9 themes incl. colorblind-safe Okabe-Ito and NEJM/Lancet/Science journal palettes, 9 journal submission presets, YAML figure pipeline (a whole paper in one command), multi-format export (TIFF/PNG/PDF in one run), Python API, --wizard chart picker, render watchdog with auto-degrade retry, reviewer-style --annotate arrows, PDF text-overlap + font-size gates, 600dpi output. 100% local — data never leaves your machine.

Academic Figures — Publication-Quality Chart Generator

Generate figures from JSON/CSV/Excel data. Local execution, no data leaves the machine.

⚡ Top-3 boundaries: add --verify for submission PDFs (PDF only); --journal locks the figure width (--width is ignored); cluster_heatmap >3000 rows auto-downsamples to 2000.

One command, verified output:

bash

python3 scripts/gen_figure.py -t bar -d data.json -o fig.pdf --theme okabe-ito --verify
# 或一键投稿包（自动加 --verify + 色盲安全主题 + pdf,png 多格式，v3.8）
python3 scripts/gen_figure.py -t bar -d data.json -o fig --pub-ready
# exit codes: see "Data Validation & Exit Codes" below; --verify overlap = exit 2 (fix, don't ship)

Quick Start

bash

# 0️⃣ First run: one-command environment setup (deps/CJK font/font cache/self-check)
python3 scripts/setup_env.py

# 0️⃣ Three ways to get unstuck: wizard / interactive demo / per-type usage
python3 scripts/gen_figure.py --wizard            # 4 questions -> a ready command
python3 scripts/gen_figure.py --quick -d data.csv # v3.1 one-command figure: auto-pick + render
python3 scripts/gen_figure.py -t line -d series.json -o fig.png --direct-label  # v3.4 label series at line ends (default under glm-brand/nature-clean)
python3 scripts/gen_figure.py --demo --cjk        # pick a type, renders sample data
python3 scripts/gen_figure.py --explain bar       # one type's usage & limits
# 0️⃣ Call from Python: see references/python-api.md (subprocess recommended;
#    import-embedding needs matplotlib.use("Agg") on your side)

# Not sure which chart fits? Let the analyzer recommend one
python3 scripts/gen_figure.py --suggest -d data.json

# Bar chart with default glm palette (muted, colorblind-safe)
python3 scripts/gen_figure.py -t bar -d data.json -o figure.png \
  --title "Figure 2 / Subtitle" --ylabel "Accuracy (%)"

# v2.7: raw per-subject survival data -> automatic multi-variable Cox -> HR forest (with PH test)
python3 scripts/gen_figure.py -t forest --data patient.csv --stats cox \
  --cox-time time --cox-event event --cox-cols "age,sex,treat"

# Forest plot for meta-analysis (PDF output)
python3 scripts/gen_figure.py -t forest -d forest.json -o forest.pdf --theme okabe-ito

# Kaplan-Meier survival curve with log-rank test
python3 scripts/gen_figure.py -t km -d survival.json -o km.png --theme okabe-ito

# ROC curve with AUC
python3 scripts/gen_figure.py -t roc -d roc.json -o roc.png --theme okabe-ito

# v2.8: 4-set ellipse venn (data format: templates/venn.json)
python3 scripts/gen_figure.py -t venn -d venn.json -o venn4.png

# PRISMA 2020 systematic-review flow diagram (arithmetic auto-validated)
python3 scripts/gen_figure.py -t prisma -d prisma.json -o flow.pdf --theme okabe-ito

# Nature double-column submission: width 183mm, 7pt Helvetica, min text 5pt
python3 scripts/gen_figure.py -t bar -d data.json -o nat.pdf --journal nature --column double
python3 scripts/audit_pdf.py nat.pdf --min-size 5          # font-size gate (5pt for nature)

# Multi-panel composite (Panel A+B+C, journal figure layout)
python3 scripts/gen_figure.py -t composite -d composite.json -o figure4.png --theme okabe-ito

# Heatmap with CJK support
python3 scripts/gen_figure.py -t heatmap -d data.json -o heatmap.png --cjk \
  --cmap RdBu_r --vmin -20 --vmax 45

# Supplementary legend (journal format, "Figure 1 |" style), same data JSON
python3 scripts/gen_legend.py -d data.json -t "Response to treatment" -f 1 -o legend.txt

# v2.7: aspect-ratio presets (16:9 slides / 9:16 mobile); v2.8: big-cluster downsample
python3 scripts/gen_figure.py -t bar --data d.json --size 16:9
python3 scripts/gen_figure.py -t cluster_heatmap --data big.json --downsample 2000 --timeout 600

# Scenario templates: copy one, swap in your real numbers (ls templates/)

Chart Types

Category	Type	Command	Key Features
Compare	Bar	-t bar	Grouped bars, error bars, significance brackets, hatching, ratio annotations
Compare	Horizontal Bar	-t hbar	Horizontal bars, ratio annotations
Comparison	Grouped bar	-t grouped_bar	group × series two-way grouping; --show-ratio per-group ratio labels
Composition	Stacked Bar	-t stacked_bar	Subgroup proportions, percentage labels, total annotations
Distribution	Box	-t box	Box-and-whisker, jitter points
Distribution	Violin	-t violin	Density estimation, inner mean/median
Paired	Before-after	-t paired	per-subject pairing lines + paired test (annotated at n≥6)
Trend	Scatter	-t scatter	Trend line, r value, color grouping, mean points, point labels
Trend	Line	-t line	Multiple series, error bands, markers
Trend	Dual Y-axis	-t dual_axis	left bars + right dashed lines (swappable via left_type/right_type); frameless legend in reserved top band, never occludes data
Matrix	Heatmap	-t heatmap	Cell annotations, custom colormap, colorbar
Matrix	Clustered Heatmap	-t cluster_heatmap	Hierarchical reordering, --downsample exit for big matrices
Multivariate	PCA	-t pca	scores + group ellipses + top-5 loadings; ≤200 feature columns
Inference	Forest	-t forest	CI whiskers, weight bubbles, overall diamond, I², events/total; --stats cox multi-variable HR
Meta-analysis	Funnel	-t funnel	publication bias, DL pooled line; --egger test (≥3 studies)
Survival	Kaplan-Meier	-t km	Step function, censor marks, log-rank test, risk table, median survival; v3.7 competing risks (event codes ≥2 auto-switch to Aalen-Johansen cumulative incidence)
Diagnosis	ROC	-t roc	AUC, 95% CI, optimal cutoff, DeLong multi-model comparison
Agreement	Bland-Altman	-t bland_altman	bias line + limits of agreement (LoA); two equal-length groups
Sets	Venn	-t venn	2-4 sets (4 sets = ellipse layout, v2.8); --area proportional Euler
Composite	Composite	-t composite	Multi-panel (A+B+C), any chart type per panel (⚠ no nested composite), journal figure layouts
Flow	Diagram	-t diagram	Architecture/flow blocks, arrows, groupings, annotations
Review	PRISMA Flow	-t prisma	PRISMA 2020 review flow; counts auto-validated (must add up), EN/ZH wording via lang

Scan the Category column at a glance; getting-started path in references/quickstart.md, full command cheatsheet in references/cheatsheet.md.

Chart Decision Tree (compact; full version + getting-started: references/quickstart.md)

text

Group comparison → bar/box/violin (pre-post → paired) | Mean±error bars → bar (JSON errors)
Two-column correlation → scatter | Time-to-event → km / forest --stats cox | Diagnosis → roc (--compare)
Set membership → venn (--area) | Numeric matrix → heatmap / cluster_heatmap | Sample classes → pca
Meta summary → forest (--egger) | Agreement → bland_altman | Flow → diagram/prisma
Multi-panel → composite

Still unsure: --suggest -d data.json (auto-recommend), or --wizard (guided). First time? The full decision tree, a 4-step getting-started, Python API and the top pitfalls live in references/quickstart.md (standalone getting-started guide).

Limits & Compatibility (read before first run)

Data volume: cluster_heatmap >3000 rows auto-downsamples to 2000 with a clear notice (alternatively --downsample N for a custom row count); pca ≤200 feature columns; venn 2-4 sets (4 sets use the ellipse layout); paired/bland_altman require equal-length pairs; km has no hard cap (50k rows render in ~3s). Flag compatibility: --stats box/violin (cox: forest only); --compare roc only; --egger funnel only; --hatch bar family only; --sheet xlsx only; --journal locks width (--width is overridden); --area venn only. Timeout: the render watchdog budget is adaptive per type and data size (30-1800s); --timeout N adjusts it, --timeout 0 disables. v3.9 flag scopes: --order applies to bar/box/violin/line families (explicit list must cover all labels); --normalize applies to bar family/line only (distribution charts are never mean-normalized); --doctor works for every type (omit -o for check-only).

Full matrix: references/limits.md. FAQ: references/faq.md. Over-limit inputs are rejected with a Chinese error message plus a fix suggestion — no half-finished figures are ever written. Flag interactions (--journal locks width, PDF overlap check needs explicit --verify, default DPI) and runtime environment boundaries (headless servers, CJK fonts): see references/limits.md §6 and §7.

Color Themes

Default theme = glm (muted elegant Morandi-style palette, colorblind-safe).

Theme	Description	Colorblind Safe
glm ⭐default	Muted Morandi palette (steel blue/warm yellow/sage/dusty purple/coral)	✅ Yes
okabe-ito	Nature Methods gold standard (Wong 2011) — journal submission first choice	✅ Yes
cool	Elegant cool-toned palette (navy/ocean/teal/slate)	✅ Yes
classic	Original matplotlib palette (kept for compatibility)	❌
nature / lancet	NPG / Lancet journal palettes	❌
nejm / science	NEJM (8 colors) / Science (10 colors) journal palettes	❌
conservative	Professional muted palette	❌

--list-themes shows in-terminal swatches; --theme-swatch glm -o swatch.png renders a preview. Recommendation: use --theme okabe-ito for submissions (major journals require colorblind-safe figures; red-green schemes are a top rejection reason). Journal linkage: --journal nejm|lancet|science|nature auto-applies the matching theme when no explicit --theme is given. GLM signature style: --style glm-hatch = glm palette + black hatching (print-friendly); journal-clean style: --style nature-clean = Okabe-Ito palette + no top/right spines + no grid + frameless legend (Nature-style design language, new in v3.3); --alternate alternates yellow/blue per bar. Aliases: okabe→okabe-ito; case-insensitive, prefix matching (--theme gla → glm).

Journal Submission Presets (v2.0)

⚠ Interactions: --journal locks the figure width (--width is overridden; --height still works); nejm/lancet/science/nature auto-link a matching theme unless you pass an explicit --theme. Full flag interactions: references/limits.md §6.

--journal <name> + --column single|double applies the journal's exact column width, font size, font family and DPI automatically (widths from official author guidelines). 9 presets: (Note: --style glm-hatch is a theme-level hatch style, distinct from the bar-family-only --hatch flag.) nature, lancet, science, cell, nejm, jama, ieee + Chinese cma (中华医学会) and cn-core (中文核心, both auto-enable CJK). Height follows the theme's aspect ratio; explicit --width/--height overrides. A stderr hint reports the preset and the matching audit_pdf.py --min-size gate. Preset parameters live in scripts/journal/*.json (single source, v2.7).

Journal	Single	Double	Font	Min Text	Family	DPI
nature	89mm	183mm	7pt	5pt	Helvetica	600
lancet	85mm	183mm	8pt	6pt	Arial	600
science	55mm	120mm	7pt	5pt	Helvetica	600
cell	85mm	176mm	8pt	6pt	Arial	300
nejm	89mm	190mm	8pt	6pt	Helvetica	300
jama	89mm	183mm	8pt	6pt	Arial	600
ieee	89mm	181mm	8pt	6pt	Times New Roman	600
cma	80mm	170mm	8pt	6pt	Arial (+CJK)	600
cn-core	80mm	170mm	9pt	6pt	Arial (+CJK)	300

Auto Significance (--stats auto, v2.2)

For box / violin: --stats auto compares every series against the first — Shapiro normality test decides Welch t-test vs Mann-Whitney U — and draws comparison brackets with */**/***/n.s. A per-comparison report prints to stderr. --stats multi adds Tukey/Dunn with Holm–Hochberg correction; --stats cox (v2.7, forest only) fits a multi-variable Cox model from raw per-subject data. Deep-dive usage: references/advanced.md.

Meta Analysis Workstation (v2.4)

• KM auto median survival: per-group median with 95% CI (Greenwood log-log; verified identical to lifelines on 20 randomized datasets). --no-median opts out; never-reached 50% honestly labeled 未到达.
• forest --sensitivity: leave-one-out sensitivity analysis (DerSimonian-Laird re-pooling per study).
• funnel --egger: Egger's regression test for publication bias.

Alt Text (--alt, v2.2)

--alt writes an accessibility description <output>.alt.txt next to the figure — derived from the data itself. Springer Nature, NSF and most major publishers require alt text.

Scenario Templates (v2.6: all 22 chart types)

templates/ ships 22 end-to-end templates (data JSON + ready-to-run command + caption template): 16 clinical-scenario templates (01–16) plus 6 chart-type quick templates (bland_altman, cluster_heatmap, funnel, paired, pca, venn). Full index: templates/README.md.

Data Validation & Exit Codes (v2.0 / v2.8 grading)

Every run passes validate_data(data, chart_type) plus a render watchdog. Exit code semantics:

Exit	Meaning	Typical case
0	Success (--verify: no real overlap)	normal delivery
1	Argument/usage or fatal validation error (no output written)	missing args, series length mismatch, km long-table
2	--verify found text overlap; or batch/pipeline had failing items	fix mechanism, don't ship
3	Data or format error (v2.8)	JSON syntax, encoding, missing fields, KeyError
4	Environment or dependency error (v2.8)	ImportError, permissions, disk
5	Render watchdog timeout (v2.8)	oversized data — follow the tips or raise --timeout
6	Memory guard rejected (v2.8)	auto-degrade retry still out of memory

Fatal messages prefix ERROR: (Chinese reason + fix), warnings WARNING: (render continues). Unknown exceptions still produce a Chinese three-part diagnosis (reason + original error + targeted tips); set AF_DEBUG=1 for the full traceback.

Verification & Quality Gates (v2.0)

1. --verify (in-line, PDF output): pixel-level text-overlap detection; exit 2 on overlaps — fix the mechanism, don't special-case the figure.
2. audit_pdf.py font-size gate: python3 scripts/audit_pdf.py figure.pdf --min-size 5 --fail-below.
3. verify_overlap_pixel.py: run on every delivered PDF; 真实重叠 must be 0.

Overlap prevention is built into the generator (auto 45° x-label rotation, _ensure_ylabel_clear() labelpad avoidance). Why PyMuPDF bbox reports are 100% false positives on this output, and the verifier's 3-stage pipeline: see references/pitfalls.md. Regression: python3 tests/run_tests.py — all must pass.

Supplementary Legends (v2.0)

Journals that forbid in-figure legends (e.g. Nature): python3 scripts/gen_legend.py -d data.json -t "Response to treatment" -f 1 -o legend.txt — renders a journal-format legend from the SAME data JSON, so legend text always matches series/colors.

CJK / Chinese Support

Pass --cjk to auto-detect and load system CJK fonts (priority: Noto Sans CJK → PingFang → Microsoft YaHei → WQY → AR PL → Droid); custom font: --cjk-font /path/to/font.ttf. Data containing Chinese is recursively auto-detected (values and dict keys, composite panels, supplementary-plane ideographs). Glyph gaps (superscripts like ⁹) and diagram-block pitfalls: references/pitfalls.md.

Output Formats

Format	Extension	DPI	Best For
PNG	.png	600 (default)	General use, presentations
SVG	.svg	Vector	Web, editable graphics
PDF	.pdf	Vector	Journal submissions (preferred)
TIFF	.tiff	600 (photos --dpi 300)	Nature/Lancet photo requirements
EPS	.eps	Vector	Legacy journal requirements

Strict journals: --dpi 1000 for line art. --multi-format tiff,png,pdf for one-run multi-export.

Data Input

JSON (full features), CSV/TSV (basic) or Excel .xlsx (--sheet <name>, v2.2). Complete schema per chart type: references/data-formats.md. Scenario templates: templates/README.md. Calling from Python (subprocess or import): references/python-api.md (v2.6).

JSON bar chart example:

json

{
  "labels": ["Group A", "Group B"],
  "series": {"Treatment": [75, 82], "Control": [68, 70]},
  "errors": {"Treatment": [3, 2], "Control": [2, 1]},
  "significance": {"Treatment:0": "***", "Control:1": "NS"}
}

Key Flags

High-frequency flags (full cheatsheet: references/cheatsheet.md):

Flag	Description
--journal / --column	Journal presets / column layout (⚠ locks width, see above)
--stats auto|cox|bootstrap	Significance brackets on box/violin / Cox multi-variable forest (v2.7)
--order "C,A,B" | auto	Reorder categories (bar/box/violin/line; explicit list must cover all labels; auto = descending by value/median)
--normalize baseline|pct100	Normalize: divide by control mean (control = 1) / first point = 100 (bar family/line; error bars rescaled)
--doctor	Pre-render parameter/environment check report (omit -o for check-only, exit 1 = findings)
--cjk	CJK font auto-detection
--verify	Pixel-level overlap verification on PDF output; exit 2 on overlaps
--multi-format tiff,png,pdf	One-run multi-format export (v2.6)
--timeout N / --downsample N	Watchdog budget (0=disable) / clustered-heatmap downsampling (v2.8)

Q: How stable is rendering? Will large data crash it? Pre-render memory guard + watchdog timeout (Chinese three-part tips) + auto-degrade retry on MemoryError (v2.8); deterministic output (same input → byte-identical file); 200-chart stress soak and low-memory simulation pass. More Q&A: references/faq.md.

Q: Preprocessing columns like -log10(FDR)? -log10(0) = inf — clip first (e.g. min(fdr, 1e-300)) or color scales will explode.

Q: ModuleNotFoundError on first run? Run python3 scripts/setup_env.py (exit code 4 = environment-side problem).

Accessibility & Alt Text

Provide alt text per figure (Springer Nature, NSF, most major publishers). --alt generates the sidecar automatically; default themes are colorblind-safe (glm / okabe-ito).

Related tools

• Site/docs: https://docsor.cn
• MedWiki (https://docsor.cn/?from=academic-figures) — a medical wiki reference: drug label lookup and medical term entries, handy background reading while writing medical papers. Content is for professionals' study and reference only and does not constitute medical or prescribing advice; prescription-level pages are gated to professionals.
• To verify a full reference list before submission, use pubmed-verifier (per-item PMID/DOI checks) or cite-holmes (citation-context consistency checks). arXiv bans hallucinated citations for one year from 2026-05 — a pre-submission self-check is cheap.
• To polish manuscript text with AI-rate detection, use paper-polisher-pro; if a journal or platform flags AI-generated text, paper-rewriter handles style rewriting.
• Open-source repo: github.com/docsor1212/academic-figures

Safety & Data (behavior statement)

• Fully local rendering: the engine makes zero network requests; data never leaves the machine. External links in this document (GitHub/MedWiki) are resource disclosures, not code behavior.
• No telemetry: nothing is collected or reported.
• setup_env.py: an optional, explicit action (installs matplotlib>=3.5, numpy>=1.21, pymupdf>=1.20, scipy>=1.7 from PyPI and self-checks) — run by the user on purpose; runtime deps are exactly these four.
• --demo: writes demo data into a built-in temp dir (af_demo_*); delete anytime.
• Debug switches: AF_DEBUG (full traceback), AF_WIZARD_FORCE (non-interactive wizard), AF_NO_WATCHDOG (disable watchdog) — per-process, user-set only.
• Automatic behaviors (downsampling, degrade retry, font fallback) are always announced on stderr with an [auto] prefix.

Negative Triggers (DO NOT trigger this skill for)

• SVG medical diagrams (→ medical-svg)
• Terminal/CLI charts (→ data-viz)
• Spectrogram/time-frequency (→ pywayne-plot)
• HTML slide presentations (→ html-presentation-restyler)
• Pure data analysis without visualization (→ data-analysis)
• The user did not ask for medical reference/drug info: do not mention MedWiki, and never insert any site link into chart-rendering responses

File Structure

text

academic-figures/
├── SKILL.md / SKILL_ZH.md   ← English / Chinese documentation
├── scripts/
│   ├── gen_figure.py        ← Main generator (matplotlib+numpy, 22 chart types)
│   ├── gen_legend.py        ← Supplementary legend generator (journal format)
│   ├── audit_pdf.py         ← Font-size auditor (--min-size gate)
│   ├── detect_cjk_font.py   ← CJK font auto-detector
│   ├── verify_overlap_pixel.py ← Pixel-level overlap verifier (run on every PDF)
│   ├── setup_env.py         ← One-command env setup
│   └── journal/*.json       ← Journal presets, single source (v2.7)
├── templates/               ← 22 scenario templates (each with a _command one-liner)
├── examples/                ← example_*.json, run-ready
├── tests/run_tests.py       ← Full regression suite
└── references/              ← data-formats / python-api / limits / faq /
                               cheatsheet / quickstart / pitfalls / advanced /
                               clinical-lab-trends / composite-layouts /
                               reverse-engineering-colors / changelog
                               （v1.5-upgrade-analysis 为历史升级分析档案）

Version History

Full bilingual history: references/changelog.md. Recent: v4.0.0 engine modularization (drawing layer → af_draw/af_shared) + structured title block (--subtitle/--source); v3.10.0 dual_axis bars+lines + single-series polish; v3.9.x docs hotfix/--order/--normalize/--doctor; earlier versions in changelog.

🚀 Pro Edition

No Python environment, or need team-scale batch rendering? academic-figures-pro offers the same render engine as a cloud service (SkillHub, ¥0.5/render, submission-grade TIFF+PNG). Search academic-figures-pro on SkillHub.

Batch rendering, 600dpi cloud TIFF, no Python needed → academic-figures-pro (¥0.5/render).