Academic Figure Prompt Generator
Reads your paper and produces top-conference-quality English image prompts for AI figure tools, covering framework, architecture, module, comparison and data-pattern diagrams.
A companion, more general academic-figure prompt generator; requires fully understanding the paper content and analyzing reference-figure color schemes before generating — a fairly rigorous process.
What does this skill do, and when should you use it?
This skill targets researchers who need publication figures for academic papers. It reads paper source files (LaTeX, Word, PDF), extracts core concepts, model architectures and math notation, then — before writing any prompt — forces a color-scheme choice among 8 preset academic palettes (including the colorblind-safe Okabe-Ito set) or user-supplied hex values. The output is an extremely detailed English prompt ready to feed into image generators like NanoBanana, Gemini, DALL-E or Midjourney. Prompts follow a strict four-layer structural template with a 13-point quality checklist, aiming at figures in the visual style of CVPR/NeurIPS-level publications.
1) Reads paper source files (LaTeX, Word, PDF) the user provides, extracting per-section concepts, methods, architectures, data flow, math symbols and tensor dimensions. 2) If a reference image is given, analyzes its palette, layout, module styling, annotation conventions and information density. 3) Before generating any prompt, presents 8 preset color schemes plus links to palette tools (Coolors, ColorBrewer, etc.) and waits for confirmation — skipped if the user already specified colors. 4) Generates English prompts, per a four-layer template (global description → section-by-section description → global annotations → style specifications), for five figure types: overall framework, network architecture, module detail, comparison/ablation, and data/behavior pattern grids — including exact hex values, dimension annotations (e.g. R^(120×6)), LaTeX-style formulas and embedded thumbnail descriptions. 5) Self-checks against a 13-item quality checklist (no empty boxes, restrained color, white-dominant, grayscale-readable) and outputs in markdown with figure number, type, palette and recommended aspect ratio. It does not generate images itself — only prompts; images come from external AI image tools.
- A grad student or postdoc with a finished CV/NLP conference paper needs an overall framework figure and network architecture diagram for the methods section but lacks professional drawing skills.
- Preparing comparison/ablation figures that place several model variants side by side with differences highlighted before a submission deadline.
- During journal revision, batch-generating figure prompts for an entire paper organized by section, with prioritization advice.
- A team with an existing reference figure wants AI-generated figures to strictly follow that figure's hex values, layout and annotation style.
- A medical/bioinformatics author wants colorblind-safe academic palettes (e.g. Okabe-Ito) and figures that stay readable in grayscale printing.
What are this skill's strengths and limitations?
- Rigorous workflow: it must understand the paper, analyze any reference image, and confirm a palette before prompting, rather than one-shotting.
- Highly structured prompts: four-layer template + exact hex values + dimension annotations + formula callouts + a thumbnail vocabulary table, far denser than freeform prompting.
- 8 built-in academic palettes (including colorblind-safe) plus 6 palette-tool links, with an anti-'AI-look' color taboo table.
- A 13-point quality checklist including a grayscale test and a no-omission rule keeps output consistent.
- Pure prompt engineering with no script dependencies — portable to any Agent Skills-compatible client; the README explicitly cites Claude Code, Gemini CLI and Cursor.
- Final image quality depends on the external image tool (NanoBanana/Gemini/DALL-E/Midjourney); the skill cannot guarantee rendering fidelity, and text/formulas in complex architecture figures are prone to typical AI-image garbling.
- The skill mandates maximum-length, maximum-density prompts, which can exceed input limits of some image tools or APIs.
- The palette-confirmation step relies on multi-turn interaction; simple clients without that capability will skip the flow.
- The repo ships no test suite or systematic evaluation for this skill — top-conference-grade output is asserted, not evidenced.
- README features (PDF parsing, 50+ palettes, image-to-image editing, SSE progress) belong to the full web platform, not this skill — the skill itself neither parses PDFs nor renders images.
How do you install this skill?
Option 1 (recommended): run npx skills add LigphiDonk/academic-figure-generator. Option 2 (manual): git clone https://github.com/LigphiDonk/academic-figure-generator.git, then copy the academic-figure-prompt folder to your client's skills directory — .gemini/skills/ for Gemini CLI, .claude/skills/ for Claude Code. Note: the repo is a monorepo containing 2 skills; this profile covers only academic-figure-prompt/SKILL.md.
How do you use this skill?
After installing, trigger it in your AI coding assistant with natural language, e.g. "Read this paper and generate figure prompts for it" or "Use the Teal+Amber palette and write a prompt for a network architecture diagram"; you can also attach a LaTeX/PDF/Word paper. If no palette is specified, the skill first shows the 8 options and palette-tool links and waits for your pick. The resulting English prompt is copied into NanoBanana / Gemini / DALL-E / Midjourney to render the figure.