30 Visual AI Skills Worth Bookmarking in 2026

If you’re new to AI skills — or specifically Claude Code skills — here’s the short version: a skill is a small markdown file plus optional scripts that teaches the agent how to do a specific task well. Skills became a breakout category in early 2026 because they distribute easier than prompts (one git clone, used forever) and solve real workflows.

This post is our opinionated tour of the 30 visual skills we’ve collected so far. Grouped by what you actually want to do, not by who made them.

Slides & decks (8)

The biggest cluster. The right skill depends on output format and audience:

anthropic-pptx — the engine. Read, edit, generate .pptx with speaker notes and editable charts. Use when the recipient will edit in PowerPoint.
theme-factory — 10 hand-tuned themes (color palette + font pairing). The post-processor for everything else.
guizang-ppt — single-file HTML, magazine-style, WebGL fluid backgrounds. By @op7418.
nanobanana-ppt — animated 4K slides + MP4 export, Nano Banana Pro powered. Also op7418.
pptgen-drawio — Draw.io-based, two modes (thesis defense + general). By 小帅同学.
ppt-creator — narrative-first (“so-what / so-now-what”). By @daymade.
slides-creator — Reveal.js / Slidev / HTML, code-friendly. Tech-talk standard.
next-slide — 53 styles across 6 families. The most stylistically diverse.

canvas-design — editorial typography, layered composition. The default for a single-image poster.
document-illustrator — turns a doc into 6 illustrated cards in 16:9 or 3:4. Perfect for 小红书.
brand-guidelines — Anthropic’s house styling (Poppins + Lora + 3 accents). Generalizable.

PDFs & documents (5)

anthropic-pdf — read, write, merge, split, watermark, encrypt, OCR. The full toolkit.
pdf-creator — Markdown → PDF with proper Chinese font support. By daymade.
anthropic-docx — Word documents end-to-end: letterheads, ToC, tracked changes.
anthropic-xlsx — spreadsheet automation, charting, data cleaning.
doc-to-markdown — the inverse: any document format → clean Markdown + extracted images.

UI & web (3)

frontend-design — distinctive, non-generic React/Tailwind output. Anti-AI-slop.
web-artifacts-builder — multi-component React with 40+ shadcn/ui pre-installed.
ui-designer — reverse: a reference image → extracted design system + implementation prompt.

Diagrams (3)

mermaid-tools — Mermaid in Markdown → high-res PNG.
excalidraw-diagram — hand-drawn vibe, editable .excalidraw JSON.
drawio-diagram — formal .drawio with style migration from a reference image.

Generative & motion (3)

algorithmic-art — p5.js, flow fields, particle systems. Outputs a sketch, not a static image.
slack-gif-creator — Slack-optimized 128×128 emoji GIFs.
cli-demo-generator — programmatic terminal recordings via VHS.

Content & writing (3)

wechat-article-writer — 9 writing styles + cover + body illustrations + voice cloning. Killer in Chinese.
humanizer-zh — strips AI signatures from Chinese text. By op7418.
youtube-clipper — URL → .mp4 / .gif / .srt for any timestamp range. By op7418.

Utilities (2)

llm-icon-finder — 100+ AI brand logos in SVG/PNG, light/dark.
capture-screen — macOS window-level screenshot automation.

How to install a skill

For Claude Code, most of these install with a single command:

git clone <skill-repo-url> ~/.claude/skills/<skill-name>

For OpenClaw or Cursor, follow each skill’s README — most are compatible across agents.

What we’re watching

Three things shape the 2026 skill landscape:

Anthropic’s official catalog — currently 17 skills, growing. Quality is high; visual coverage is partial.
Chinese designer-authors — @op7418 and @xstongxue are shipping skills that feel like products, not toy projects.
Cross-agent compatibility — most new skills target Claude Code + OpenClaw + Cursor simultaneously. Lock-in is decreasing.

If you only install three from this list: theme-factory (style layer), canvas-design or guizang-ppt (output layer), and humanizer-zh (if you write in Chinese). Everything else builds on top.