Gemini Images

AI image generation skill for Claude Code - Generate consistent game sprites, character sheets, and visual assets using Gemini's spatial understanding and reference sheet methodology.

Gemini can preserve visual features across different views, poses, and compositions when given specific image references. This skill leverages that capability through precise instructions for managing your assets:

• Character sheets - Generate multiple views (front/back/side) that establish a consistent visual identity
• Reference composition - Combine existing images with new elements while preserving original features
• Structured prompts - Use explicit labels and visual anchors to maintain consistency across iterations
• Small deltas - Change one thing at a time (pose, accessory, lighting) to preserve everything else

See skills/gemini-images/SKILL.md for complete methodology and techniques.

Use in Claude Code

This is a Claude Code skill. Install it from the marketplace:

/plugin marketplace add krasnoperov/claude-plugins
/plugin install gemini-images@krasnoperov-plugins

Once installed, use the /gemini-images skill in your conversations:

/gemini-images generate a pixel art character sheet with front, back, and side views
/gemini-images edit character.png "add armor and sword"
/gemini-images compose character.png background.png "place character in forest scene"

Command Line Usage

You can also use this package directly via npx:

export GEMINI_API_KEY="your-key-here"

# Generate from text
npx -y @krasnoperov/gemini-images@latest generate "pixel art tree" --output tree.png

# Transform image
npx -y @krasnoperov/gemini-images@latest edit tree.png "add glowing runes" --output tree-magic.png

# Combine references
npx -y @krasnoperov/gemini-images@latest compose hero.png sword.png "character holding sword" --output hero-armed.png

Get your API key: Google AI Studio

Core Operations

generate "<prompt>"                          Text → Image
edit <image> "<prompt>"                      Image + Instructions → Image
compose <img1> <img2> ... "<prompt>"         Images + Instructions → Image

These three operations compose into any workflow you need.

Examples

See skills/gemini-images/examples/ directory:

1. 01-primitives.sh - Basic usage of all three operations
2. 02-consistency-pipeline.sh - Reference sheet methodology step-by-step

Reference Sheet Methodology

Gemini has spatial understanding - it preserves visual features when given specific references.

Character Sheets

npx -y @krasnoperov/gemini-images@latest generate \
  "Character sheet: front view, back view, side view. Character: ranger, auburn hair, green jerkin." \
  --output character_sheet.png

Structured Prompts

npx -y @krasnoperov/gemini-images@latest compose character_sheet.png accessories.png \
  "Image 1: Character sheet
Image 2: Accessories
Character: From image 1, front-facing
Items: Backpack from image 2
Lighting: Soft natural light, top-left
Camera: Eye level, medium shot" \
  --output scene.png

Options

--model <model>           gemini-3-pro-image-preview (default) or gemini-2.5-flash-image
--aspect-ratio <ratio>    1:1 (default), 16:9, 9:16, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 21:9
--image-size <size>       1K (default), 2K, 4K (Pro model required for 2K/4K)
--output <path>           Output file or directory (default: ./output/)
--api-key <key>           Gemini API key (or set GEMINI_API_KEY env var)

Model Selection

| Feature | Flash | Pro | |---------|-------|-----| | Max Resolution | 1K | 4K | | Reference Images | 1 | 14 | | Speed | Fast | Moderate | | Best For | Quick iterations | Production assets |

Development

npm run build      # Build TypeScript
npm run typecheck  # Type checking
npm run dev        # Dev mode with type stripping

Acknowledgments

Reference sheet methodology based on Towards Data Science.

License

MIT License - Copyright (c) 2025 Aleksei Krasnoperov

See LICENSE file for details.