Designed images from a regular code-capable LLM—no image-generation model required.
Illustrator is a lightweight, browserless Agent Skill that turns natural-language direction, content, code, and supplied images into designed visuals through local rendering. Start from the bundled examples or define a visual style entirely your own.
Project article (Chinese): Design rationale, usage, and real-world examples
- No image-generation model — a regular code-capable LLM can create visuals by generating local rendering code.
- Lightweight and browserless — no Chromium, Puppeteer, or browser runtime.
- Examples included, styles unrestricted — use the bundled examples as references or bring your own visual direction.
- Code-driven and locally rendered — visuals are produced through generated rendering code rather than a black-box image model.
- Editorial illustrations and posters
- Article and social media visuals
- Photo-led compositions using supplied images
- Data-led reports and information graphics
- Code snippets, windows, and diffs
- Fully custom visual systems
Requires Node.js 20 or newer and an Agent Skills-compatible coding agent.
npx skills add Visualizeit/illustrator --skill illustratorAfter installation, describe the visual you want to your agent:
/illustrator Create a 1600×900 editorial illustration about focus and deep work.
/illustrator Redesign this content using the bundled Flash Diary example as a visual reference.
/illustrator Turn this code into a clean, presentation-ready image.
Below are published article visuals created with Illustrator.
Related article: 不用生图模型，我让 Agent “画”出了这些图 (Chinese)
Bundled examples include their rendering source, assets, and output previews, making them easy to inspect, modify, and render again.
A bright, photo-led city diary built from supplied imagery, bold type, and compact editorial details.
A colorful market poster combining oversized typography with crisp geometric flowers.
A contemporary specimen study shaped by natural forms, measured annotations, and catalog metadata.
A typographic specimen that turns the ILLUSTRATOR wordmark into its own measured, color-coded construction.
A restrained motorsport editorial combining motion photography with compact performance data.
A landscape automation visual built around a semantic spreadsheet and one precise interaction.
An analytical report visual using crisp comparison lines and overlapping translucent fields.
A minimal color-study poster with one dominant field and concise editorial typography.
A presentation-ready split diff with syntax highlighting and clear added and removed regions.
A light code window with compact file chrome and readable syntax highlighting.
A minimal dark code image for documentation, articles, and social posts.
Copyright 2026 Visualizeit. Licensed under the Apache License 2.0; redistributions must retain the required license and attribution notices in NOTICE.
Using Illustrator does not by itself require attribution on generated images. Bundled fonts and example photographs remain under their respective terms; see THIRD_PARTY_NOTICES.md.
Requires Node.js and pnpm.
pnpm install
pnpm run validateThe Skill source lives in skills/illustrator. Noto Sans SC and JetBrains Mono are bundled for consistent text and code rendering; see THIRD_PARTY_NOTICES.md for font and image sources.
Important
Public alpha. Illustrator is usable today, but its Skill interface and runtime packaging may change before v1.0.