Pixel Art Generator

Pixel Art Generator

Generate pixel art PNGs from text descriptions. You design the art yourself and render it with the bundled Python script.

Prerequisites

Before generating pixel art, ensure:

1. Pillow Python package is installed: pip install Pillow

If Pillow is missing, install it before proceeding.

Design Process

When a user requests pixel art, follow these steps:

Step 1: Choose Grid Size

Pick a grid size based on subject complexity:

| Complexity | Grid | Examples | |------------|------|----------| | Simple icons | 8x8 | heart, star, arrow, smiley | | Standard sprites | 16x16 | character, animal, item | | Detailed scenes | 32x32 | landscape, building, vehicle |

If the user specifies a size (e.g., "10x10"), use that instead.

Step 2: Choose a Palette

Constrain yourself to 4-8 colors. Fewer colors = better pixel art. Pick colors that suit the subject. See ${CLAUDE_PLUGIN_ROOT}/skills/pixel-art-gen/references/pixel-art-guide.md for suggested palettes.

Step 3: Design in Layers

Think through the design systematically:

1. Silhouette — outline the main shape
2. Base fill — fill with primary colors
3. Details — add features (eyes, patterns, textures)
4. Shading — add darker/lighter variants for depth

Step 4: Generate JSON Pixel Data

Create a JSON object using the sparse coordinate format. Only list non-background pixels:

{
  "width": 8,
  "height": 8,
  "background": "#FFFFFF",
  "grid_lines": false,
  "pixel_size": 32,
  "pixels": [
    {"x": 3, "y": 0, "color": "#FF0000"},
    {"x": 4, "y": 0, "color": "#FF0000"}
  ]
}

Important rules:

• Coordinates are 0-indexed. x is column (left to right), y is row (top to bottom).
• Only include pixels that differ from the background color.
• Use hex colors (e.g., "#FF6B35") for precision. CSS named colors (e.g., "red") and "transparent" are also supported.
• Double-check coordinates — off-by-one errors ruin pixel art.

Step 5: Render to PNG

1. Write the JSON data to a file (e.g., pixel_art.json)
2. Run the render script:

python3 "${CLAUDE_PLUGIN_ROOT}/skills/pixel-art-gen/scripts/render_pixel_art.py" \
  pixel_art.json \
  -o pixel_art.png

1. Display the result to the user using the Read tool on the output PNG file.

Script arguments:

| Argument | Required | Default | Description | |----------|----------|---------|-------------| | input | Yes | - | JSON file path, or - for stdin | | -o, --output | No | pixel_art.png | Output PNG file path | | -p, --pixel-size | No | 32 | Size of each logical pixel in the output | | -g, --grid-lines | No | off | Enable 1px grid lines at pixel boundaries | | --no-grid-lines | No | - | Explicitly disable grid lines (overrides JSON) |

Step 6: Iterate

After displaying the result, offer to adjust:

• Colors or palette
• Individual pixel positions
• Grid size (start over at larger/smaller resolution)
• Add/remove grid lines
• Change background color

Workflow Summary

When a user requests pixel art:

1. Determine the subject and appropriate grid size
2. Choose a constrained color palette (4-8 colors)
3. Mentally design the art layer by layer
4. Generate the JSON pixel data with explicit x,y coordinates
5. Write the JSON to a file
6. Run the render script to produce a PNG
7. Display the PNG to the user with the Read tool
8. Offer to make adjustments

Tips for Better Pixel Art

• Symmetry: Many subjects (faces, creatures, objects) benefit from horizontal symmetry
• Outline: A 1px dark outline helps shapes read clearly at small sizes
• Anti-aliasing: For diagonal lines, use intermediate colors at stair-step edges
• Highlights: Place lighter colors on top-left to suggest a light source
• Shadows: Place darker colors on bottom-right

Troubleshooting

| Issue | Solution | |-------|----------| | Pillow not installed | Run pip install Pillow | | Garbled output | Check x,y coordinates are within grid bounds (0 to width-1, 0 to height-1) | | Colors look wrong | Use hex codes for precision; named colors may vary | | Image too small/large | Adjust -p pixel size (default 32) |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: thejacedev
• Source: thejacedev/claude-code-skills
• License: MIT

Install and usage instructions live in the source repository linked above.