name: gemini-image
description: Generate images using Google Gemini and Imagen models via scripts/. Use for AI image generation, text-to-image, creating visuals from prompts, generating multiple images, custom aspect ratios, and high-resolution output up to 4K. Triggers on "generate image", "create image", "imagen", "text to image", "AI art", "nano banana".
license: MIT
version: 1.0.0
keywords: image generation, imagen, gemini-3-pro, gemini-2.5, text-to-image, AI art, nano banana, 4K resolution, aspect ratio

Gemini Image Generation

Generate high-quality images from text prompts using Google's Gemini and Imagen models through executable scripts.

When to Use This Skill

Use this skill when you need to:
- Create visual content from text descriptions
- Generate multiple image variations
- Create images at specific resolutions (1K, 2K, 4K)
- Produce images for different aspect ratios (social media, banners, etc.)
- Generate photorealistic images or artistic visuals
- Create images with person generation controls
- Batch generate multiple images at once
- Combine with text generation for complete content creation

Available Scripts

scripts/generate_image.py

Purpose: Generate images using Gemini 3 Pro Image or Imagen 4 models

When to use:
- Any image generation task
- Multiple image generation (1-4 per request)
- Custom resolution and aspect ratio needs
- Professional asset creation
- Photorealistic or artistic image generation

Key parameters:
| Parameter | Description | Example |
|-----------|-------------|---------|
| prompt | Text description (required) | "A futuristic city at sunset" |
| --model, -m | Model to use | gemini-3-pro-image-preview |
| --output-dir, -o | Output directory for images | images/ |
| --name, -n | Base name for output files | artwork |
| --no-timestamp | Disable auto timestamp | Flag |
| --aspect, -a | Aspect ratio | 16:9 |
| --size, -s | Resolution | 2K or 4K |
| --num | Number of images (1-4) | 4 |
| --person | Person generation policy | allow_adult |

Output: List of saved PNG file paths

Workflows

Workflow 1: Basic Image Generation

python scripts/generate_image.py "A futuristic city at sunset with flying cars"

• Best for: Quick image generation, prototypes
• Model: gemini-3-pro-image-preview (default, highest quality)
• Output: images/generated_image_YYYYMMDD_HHMMSS.png

Workflow 2: Social Media (Instagram, Facebook)

python scripts/generate_image.py "Minimalist coffee shop interior" --aspect 1:1 --size 2K --name coffee-shop

• Best for: Instagram posts, profile pictures
• Aspect: 1:1 (square format)
• Resolution: 2K (2048x2048)
• Output: images/coffee-shop_YYYYMMDD_HHMMSS.png

Workflow 3: YouTube Thumbnails (16:9)

python scripts/generate_image.py "Tech gadget review thumbnail with vibrant colors" --aspect 16:9 --size 2K --name thumbnail

• Best for: YouTube, video thumbnails
• Aspect: 16:9 (widescreen)
• Resolution: 2K (2752x1536)
• Output: images/thumbnail_YYYYMMDD_HHMMSS.png

Workflow 4: Multiple Variations

python scripts/generate_image.py "Abstract geometric patterns in blue and gold" --num 4 --name abstract

• Best for: A/B testing, design options
• Generates: 4 distinct variations
• Output: images/abstract_YYYYMMDD_HHMMSS_0.png, images/abstract_YYYYMMDD_HHMMSS_1.png, etc.

Workflow 5: Custom Output Directory

python scripts/generate_image.py "Detailed architectural rendering of modern museum" --aspect 16:9 --size 4K --output-dir ./professional/ --name museum

• Best for: Print materials, high-end assets, organized projects
• Model: gemini-3-pro-image-preview only (for 4K)
• Resolution: 4K (5504x3072 for 16:9)
• Directory created automatically if it doesn't exist

Workflow 6: Photorealistic Images (Imagen 4)

python scripts/generate_image.py "Robot holding a red skateboard in urban setting" --model imagen-4.0-generate-001 --aspect 16:9 --size 2K --num 2 --name robot-skate

• Best for: Realistic photos, product shots
• Model: imagen-4.0-generate-001 (photorealistic)
• Notes: English prompts only
• Max 4 images per request

Workflow 7: Blog Post Featured Image

python scripts/generate_image.py "Serene mountain lake at sunrise with reflections" --aspect 16:9 --size 2K --output-dir ./blog-images/ --name featured-image

• Best for: Blog headers, article images
• Combines well with: gemini-text for blog content generation

Workflow 8: Content Creation Pipeline (Text + Image)

# 1. Generate content (gemini-text skill)
python skills/gemini-text/scripts/generate.py "Write a product description for smart home device"

# 2. Generate product image (this skill)
python scripts/generate_image.py "Sleek modern smart home device on white background" --aspect 4:3 --size 2K --name product

# 3. Create social media post

• Best for: E-commerce, marketing campaigns
• Combines with: gemini-text, gemini-batch for batch production

Workflow 9: Disable Timestamp

python scripts/generate_image.py "Fixed filename image" --name my-image --no-timestamp

• Best for: When you want complete control over filename
• Output: images/my-image.png (no timestamp)
• Use when: Generating files for specific naming schemes or automated pipelines

Parameters Reference

Model Selection

Aspect Ratios

Note: 4K resolution only available with gemini-3-pro-image-preview

Resolution Guide

Person Generation Policy

Output Interpretation

File Naming

• Default format: {name}_YYYYMMDD_HHMMSS.png (auto timestamp)
• Single image example: artwork_20260130_031643.png
• Multiple images: {name}_YYYYMMDD_HHMMSS_0.png, {name}_YYYYMMDD_HHMMSS_1.png, etc.
• Without timestamp (--no-timestamp): {name}.png
• Script prints: "Saved: /path/to/file.png"

Image Quality

• All images include SynthID watermark for authenticity
• PNG format for lossless quality
• Can be converted to JPEG/WEBP if needed
• 4K images are significantly larger file sizes

Error Messages

• "Model not available": Check model name spelling
• "Unsupported size": Verify size/model combination
• "Aspect ratio error": Use supported ratios for selected model

Common Issues

"google-genai or pillow not installed"

pip install google-genai pillow

"Image generation failed"

• Check prompt length (too verbose can fail)
• Try simpler, more focused prompts
• Verify model availability in your region
• Check API quota limits

"Unsupported aspect ratio"

• Check if ratio is supported by selected model
• Imagen 4 has fewer ratio options than Gemini
• Use 16:9 or 1:1 for best compatibility

"4K not supported"

• 4K only works with gemini-3-pro-image-preview
• Use --size 2K for other models
• Try --model gemini-3-pro-image-preview --size 4K

"Imagen prompt language error"

• Imagen models support English prompts only
• Use gemini-3-pro-image-preview for other languages
• Translate prompt to English for Imagen

File too large for storage

• Use --size 1K for smaller files
• Compress images after generation
• Convert PNG to JPEG for web use

Best Practices

Prompt Engineering

• Be specific and descriptive
• Include style descriptors (e.g., "photorealistic", "digital art")
• Mention lighting, mood, and composition
• Use analogies for complex concepts
• Avoid negative prompts (describe what you want, not what to avoid)

Model Selection

• Use gemini-3-pro-image-preview for: High quality, text rendering, 4K
• Use gemini-2.5-flash-image for: Speed, high volume
• Use imagen-4.0-generate-001 for: Photorealism, product shots

Performance Optimization

• Generate multiple images at once with --num
• Use lower resolution for previews
• Batch requests for high-volume needs (gemini-batch skill)
• Cache results for repeated requests

Quality Tips

• Use 2K resolution for most web uses
• 4K only when maximum detail is needed
• Combine specific prompts with style guidance
• Test prompts with --num 1 before generating batches

Cost Management

• Use flash models for cost efficiency
• 4K generation costs significantly more
• Batch multiple requests when possible
• Generate at 1K for testing, 2K/4K for final

Related Skills

• gemini-text: Generate text content alongside images
• gemini-tts: Create audio for image-based content
• gemini-batch: Process multiple image requests efficiently
• gemini-embeddings: Generate image embeddings for similarity search

Quick Reference

# Basic
python scripts/generate_image.py "Your prompt"

# Social media (1:1)
python scripts/generate_image.py "Prompt" --aspect 1:1 --size 2K --name social-post

# YouTube thumbnail (16:9)
python scripts/generate_image.py "Prompt" --aspect 16:9 --size 2K --name thumbnail

# 4K high quality
python scripts/generate_image.py "Prompt" --aspect 16:9 --size 4K --name high-res

# Multiple variations
python scripts/generate_image.py "Prompt" --num 4 --name variations

# Custom directory
python scripts/generate_image.py "Prompt" --output-dir ./my-images/ --name custom

# Photorealistic
python scripts/generate_image.py "Prompt" --model imagen-4.0-generate-001 --aspect 16:9 --size 2K --name photo

# No timestamp
python scripts/generate_image.py "Prompt" --name fixed-name --no-timestamp

Reference

• See references/ for model documentation (if available)
• Get API key: https://aistudio.google.com/apikey
• Documentation: https://ai.google.dev/gemini-api/docs/image-generation
• SynthID: https://deepmind.google/technologies/synthid/