Zum Inhalt springen

Skills / yoyopixel

yoyopixel

Pixel art without GPU. Any text-only LLM can draw. Size + prompt → self-contained HTML. Claude Code skill / MCP compatible.

1von @SbNamevor 13d aktualisiertMITGitHub →

Installation

Kompatibilitaet

Claude CodeGemini

Beschreibung


What is YoYoPixel?

Text-only LLMs can generate images — no GPU, no DALL-E, no multimodal model needed.

YoYoPixel is a skill/prompt framework that teaches any LLM (Claude, GPT, Gemini, Llama, Qwen...) to generate pixel art as pure HTML/CSS/JS code. The LLM writes structured text data, the browser renders it as pixel art.

Traditional: prompt → GPU → image model → PNG
YoYoPixel:   prompt → any text LLM → HTML code → pixel art in browser
/pixelart 32x32, a wizard with a glowing staff

The AI auto-decides everything: palette, detail level, rendering method, shading, animations, atmosphere. No configuration needed.

Capabilities at a glance

| Size Range | What it can do | Rendering | |-----------|----------------|-----------| | 8×8 | Icons, emojis, game items (sword, potion, key, chest) | CSS box-shadow | | 12–16px | Character sprites, avatars, simple creatures | CSS Grid + JS | | 24–32px | Detailed characters with outfits, weapons, accessories | CSS Grid + JS | | 48px+ | Complex illustrations with sub-pixel shading | CSS Grid + JS | | 48–96px | Game assets, buildings, animals, tilesets | Procedural texture engine | | 96–256px | Full procedural landscapes, cityscapes, world landmarks | Canvas + FBM noise |


Install as Claude Code Skill

Option 1: Project-level (recommended)

# Clone
git clone https://github.com/SbName/yoyopixel.git

# Copy skill into your project
cp -r yoyopixel/.claude/skills/pixelart YOUR_PROJECT/.claude/skills/

Option 2: Global (all projects)

cp -r yoyopixel/.claude/skills/pixelart ~/.claude/skills/

Use it

/pixelart 16x16, a fire mage with staff
/pixelart 32x32, cyberpunk cat with neon goggles
/pixelart 192x128, mountain sunset with lake reflection
/pixelart 24x24, ancient Chinese swordsman in moonlight
/pixelart 80x85, stone cottage with red tile roof
/pixelart 50x40, a fox sitting in grass

# Tileset mode — generates 25-40 related assets on one sheet
/pixelart tileset, medieval village
/pixelart tileset, dungeon crawler

Two input modes: size + prompt for single assets, tileset + theme for full sprite sheets.

Claude auto-decides everything based on size and prompt:

  • Size determines detail level, max colors, and shading complexity
  • Prompt content determines rendering method (CSS Grid, Canvas landscape, or procedural texture engine)
  • Keywords trigger auto-animations (weapon → gleam, night → stars, fire → embers)
  • Subject type selects texture algorithms (building → stone/tiles, animal → fur/scales/feathers)

Showcases

Live Demo — try all demos online, no download needed.

| Page | Link | |------|------| | All-in-one Showcase | sbname.github.io/yoyopixel/showcase.html | | Landscape Generator | sbname.github.io/yoyopixel/landscape.html | | World Wonders | sbname.github.io/yoyopixel/wonders.html | | Stone Cottage | sbname.github.io/yoyopixel/prototype-cottage.html | | Style Gallery | sbname.github.io/yoyopixel/prototype-gallery.html | | Animals | sbname.github.io/yoyopixel/prototype-animals.html |

Characters & Scenes

Procedural Generators

Procedural Texture Engine (NEW)

Game-ready assets using region-based procedural textures — LLM declares regions + materials, engine auto-fills textures:

19 procedural texture algorithms + sprite template system with per-material hue-shift shading and adaptive outlines

Full tileset demo:

Multi-Size Gallery


How It Works

Small art (≤48px): String-array pixel data

Each character in a string = one pixel. Human-readable, AI-writable.

const wizard = {
  width: 24, height: 24,
  palette: {
    '.': 'transparent',
    'H': '#2a1a4e',     // hat
    'S': '#f5d0a0',     // skin
    'R': '#6a3cbf',     // robe
  },
  pixels: [
    '........HHHH............',
    '......HHHHHHHH..........',
    '....HHHHHHHHHHH.........',
    '...HHHSSSSSSSHHH........',
  ]
};

Rendered via CSS Grid (per-pixel animation) or CSS box-shadow (zero JS, pure CSS).

Medium art (48–96px): Procedural texture engine

For game assets, buildings, creatures — the LLM declares regions + materials, the engine fills textures automatically:

Region-based: { shape: "rect", bounds: [12,47,68,74], material: "stone_wall" }
12 texture types: stone, tiles, wood, bark, foliage, fur, scales, feathers, spiral...
Auto post-processing: directional shading → edge outline → local shadows → glow

No per-pixel specification needed for surfaces. One engine generates cottages, towers, trees, foxes, owls, treasure chests — all with consistent style.

Large art (≥64px): Canvas-based procedural generation

FBM Noise → Terrain shape
Ridge Noise → Mountain ridgelines
Bayer 4×4 Dithering → Palette-limited gradients
Layered Rendering → Sky → Stars → Clouds → Mountains → Trees → Water → Mist

All algorithms run in-browser. No server, no API calls.


For AI Platforms

Claude Code

Install the skill, then:

/pixelart 32x32, a knight in silver armor

MCP / OpenClaw

{
  "size": "32x32",
  "prompt": "a knight in silver armor"
}

What the AI auto-decides

| Input | AI figures out | |-------|---------------| | Size 8x8 | minimal detail, 3-5 colors, CSS box-shadow, no animation | | Size 32x32 | detailed, 12-20 colors, CSS Grid, breathe + gleam + fade-in | | Prompt mentions "night" | adds stars, moon, mist atmosphere | | Prompt mentions "fire" | adds ember particles, glow effects | | Prompt mentions "sword" | adds gleam animation on weapon pixels | | Size 192x128 | canvas mode, FBM noise, procedural generation | | Prompt tileset, village | tileset mode, ~35 assets, procedural textures, shared palette, single sheet |

Full decision rules → pixelart-skill.md


Tech Stack

| Component | Technology | |-----------|-----------| | Rendering (small) | CSS Grid / CSS box-shadow | | Rendering (medium) | Canvas + Procedural texture engine | | Rendering (large) | Canvas + FBM noise + PixelBuffer | | Terrain generation | FBM (Fractal Brownian Motion) noise | | Mountain ridges | Ridge noise (abs-inverted FBM) | | Gradient dithering | Bayer 4×4 ordered dithering | | Surface textures | 19 types: stone, tiles, wood, bark, foliage, fur, scales, feathers, spiral, cloth, woven, metal, crystal, bone, rope, coral, stucco, planks, smooth | | Post-processing | Multi-pass: ambient occlusion → directional light → specular highlights → edge outline | | Composition | RGBA pixel buffer, alpha-aware blit with contact shadows, assembly pattern for complex assets | | Animation | CSS @keyframes + requestAnimationFrame | | Atmosphere | Pure CSS particles (rain, petals, embers, snow) | | Output | Self-contained HTML (zero external dependencies) |


Project Structure

pixelai/
├── .claude/skills/pixelart/
│   └── SKILL.md              ← Claude Code skill definition
│
├── README.md                 ← You are here
├── pixelart-skill.md         ← Full skill reference (AI decision rules)
├── GUIDE.md                  ← Technical implementation guide
├── LICENSE                   ← MIT
│
├── showcase.html             ★ All-in-one demo page
├── landscape.html            ★ Procedural landscape generator
├── wonders.html              ★ World wonders (8 landmarks)
├── index.html                ← Multi-size gallery
├── swordsman.html            ← Moonlit swordsman scene
├── cyberpunk.html            ← Neon cyberpunk scene
├── dragon.html               ← Fire dragon scene
├── icons.html                ← 8×8 icon collection
│
├── prototype-cottage.html    ★ Procedural texture engine: stone cottage + tileset
├── prototype-gallery.html    ★ Style gallery: tower, sakura tree, chest, well
├── prototype-animals.html    ★ Organic textures: fox, koi, owl, snail
├── tileset-village.html      ★ Full village tileset (~35 assets on one sheet)
├── tileset-tavern.html       ★ Tavern interior tileset (~45 assets)
├── engine-v2.html            ★ Engine v2: cloth, metal, crystal, bone, assembly
└── engine-v3.html            ★ Engine v3: sprite templates, hue-shift shading, adaptive outline

License

MIT — use it however you want.

Aehnliche Skills

yoyopixel | hub.ai-engineering.at