Academic Figure Prompt — Modern ML Airy Style
Reads your paper and produces extremely detailed English prompts for AI image tools, yielding academic figures in the pastel aesthetic of recent ICLR/NeurIPS/ICML papers.
Generates prompts for AI image tools to produce academic figures — a pure text-generation task, with rules stated as validated across 7 iterations, no execution risk.
What does this skill do, and when should you use it?
This is a prompt-generation skill for academic paper figures, version 4.0.0, located at academic-figure-prompt-pastel/SKILL.md within the LigphiDonk/academic-figure-generator repository (which bundles two separate skills). It does not render images itself; instead it reads your paper and writes exhaustive English prompts for you to feed into any AI image generator. The aesthetic — pure white canvas, white panels with soft drop shadows, rounded Nunito/Poppins fonts, pastel token squares, colored keywords, and pill-shaped concept labels — was refined over 7 iterations to match top-tier ML conference figures from 2024–2025. The parent repo is a full FastAPI + React platform released under MIT; this skill is the standalone agent-skill slice of it.
The skill runs a three-step workflow: first it reads the paper's source files to extract core concepts, methods, and data flow, identify figure placement, and understand math notation; second it presents four color schemes (Warm ML, Cool Research, Earthy Warm, or custom) and waits for user confirmation; third it generates a full English prompt in three layers — global description, per-panel descriptions, and style specifications — covering token squares with exact hex fills and 1px borders, curve thumbnails, formulas, inline illustrations, decision trees, arrows, and pill labels, followed by a final quality checklist. Its declared tool is bash; declared stages are writing, research, and review.
- An ML/RL researcher submitting to ICLR/NeurIPS/ICML who needs prompt text for a method framework or architecture diagram
- A graduate student revising a paper who must add ablation-comparison or module-detail figures consistent with recent conference aesthetics
- A researcher with clear ideas but weak drawing skills who wants AI image tools to do the rendering
- A content author preparing multiple stylistically consistent diagrams for a technical report or survey
- A lab with specific palette needs (colorblind-friendly, journal style) that wants to define a custom token color scheme
What are this skill's strengths and limitations?
- Style rules validated over 7 iterations and specified down to hex values, font sizes, shadow parameters, and pixel spacing — highly reproducible output
- English prompts (best for image tools) with Chinese interaction text — friendly to Chinese-speaking researchers
- Built-in color, element, and layout vocabularies covering framework, architecture, comparison, and module diagrams
- Layout is content-driven and asymmetric by design, avoiding cookie-cutter 2×2 grids
- Pure-text skill with zero runtime dependencies, MIT licensed, freely modifiable
- Generates prompts only, not images — you must separately arrange and pay for an AI image tool, and final quality depends on how well that tool follows long prompts
- Tightly bound to one aesthetic (modern pastel ML conference style); unsuitable for journals demanding plain line art or other conventions
- SKILL.md defines 4 color schemes (P1–P4), which conflicts with the README's platform-level claims of 50+ or 8 schemes — don't conflate them
- No test suite, sample output images, or adoption evidence in the source material; actual rendering quality can't be verified from documentation
How do you install this skill?
The repo supports one-line install via npx skills add LigphiDonk/academic-figure-generator (note: the README's manual-install examples copy the sibling skill folder academic-figure-prompt; for this skill the folder is academic-figure-prompt-pastel — copy that folder instead): into .claude/skills/ for Claude Code or .gemini/skills/ for Gemini CLI. It is a single SKILL.md with no extra dependencies or API keys (the image tool you later use is arranged separately).
How do you use this skill?
After installation, trigger it in conversation with your AI coding assistant, e.g. "Read this paper and generate a prompt for a modern ML paper figure", or use the documented trigger phrases "pastel风格论文配图" / "modern ML figure prompt". The assistant analyzes the paper, shows color schemes P1–P4 for you to pick, then outputs a Markdown block with the complete English prompt (figure title, diagram type, recommended resolution). Paste the prompt into any AI image tool to render.