Skip to main content
The generate_image tool creates high-quality photorealistic images using an advanced AI image generation model.

​
generate_image

Generate or edit images using Gemini 3 Pro Image (Nano Banana Pro).

​
Input Parameters

​
prompt
string
required
Image description or edit instructions. The prompt is automatically enhanced with professional photography and rendering techniques.
​
resolution
string
default:"1K"
Output resolution:
  • 1K - Standard resolution (1024px)
  • 2K - High resolution (2048px)
  • 4K - Ultra-high resolution (4096px)
​
aspect_ratio
string
default:"1:1"
Image aspect ratio:
  • 1:1 - Square
  • 3:4 - Portrait (standard photo)
  • 4:3 - Landscape (standard photo)
  • 9:16 - Vertical (mobile/story)
  • 16:9 - Widescreen (desktop/video)

​
Response

Returns image as base64-encoded data URI.
​
image
string
Data URI with format: data:image/png;base64,{base64_data}For LLM: Full data URI prefixed with IMAGE_GENERATED:For User: Confirmation message: I've generated the image based on your prompt: "{prompt}"

​
Usage Examples

Basic generation: 
{
  "prompt": "A serene mountain landscape at sunset"
}
High resolution portrait: 
{
  "prompt": "Professional headshot of a software engineer",
  "resolution": "2K",
  "aspect_ratio": "3:4"
}
Widescreen scene: 
{
  "prompt": "Futuristic city skyline with flying cars",
  "resolution": "4K",
  "aspect_ratio": "16:9"
}

​
Prompt Enhancement

User prompts are automatically enhanced with professional art direction to ensure high-quality photorealistic output: Input prompt: 
"A cat sitting on a windowsill"
Enhanced prompt sent to model: 
Subject: A cat sitting on a windowsill.
Camera Gear: Leica M11, Hasselblad X2D, 85mm Prime f/1.2.
Lighting: Rembrandt lighting, volumetric god rays, high-key studio, cinematic rim light.
Aesthetic: Photorealistic, 8k resolution, highly detailed skin texture, soft bokeh background, subsurface scattering, Octane Render, Ray-traced global illumination.
Composition: Minimalist, strong central subject, purposeful negative space, high-end editorial and cinematic standards.
Style: Avoid generic "AI-style" neon/purple tropes. No purple-blue gradients. No colored shadows.

User Prompt: "A cat sitting on a windowsill"

​
Enhancement Features

Camera equipment simulation:
  • Professional camera bodies (Leica M11, Hasselblad X2D)
  • Premium lens characteristics (85mm Prime f/1.2)
  • Shallow depth of field and bokeh
Lighting techniques:
  • Rembrandt lighting for dramatic shadows
  • Volumetric god rays for atmospheric depth
  • High-key studio lighting options
  • Cinematic rim lighting
Rendering quality:
  • 8K resolution detail level
  • Subsurface scattering for realistic skin/materials
  • Octane Render quality standards
  • Ray-traced global illumination
Composition principles:
  • Minimalist aesthetic
  • Strong central subject focus
  • Purposeful negative space
  • Editorial and cinematic standards
Style constraints:
  • Avoids generic “AI art” aesthetics
  • No purple/blue gradients or neon colors
  • No colored shadows
  • Natural, photorealistic output

​
File Handling

Temporary storage:
  • Generated images are saved to: {workspace}/generated/gen-{timestamp}.png
  • Timestamp format: 20060102-150405 (YYYYMMDD-HHMMSS)
  • Files are automatically deleted after conversion to data URI
Example filename: 
/workspace/generated/gen-20260301-143022.png

​
Error Conditions

  • Image generation failed: {error} - Model API error or generation failure
  • Failed to read generated image: {error} - File system error after generation

​
Implementation Details

Generation script:
  • Uses Python UV runtime for dependency management
  • Script location: /usr/lib/node_modules/openclaw/skills/nano-banana-pro/scripts/generate_image.py
  • Communicates with Gemini 3 Pro Image API
Command structure: 
uv run /path/to/generate_image.py \
  --prompt "enhanced prompt" \
  --filename /path/to/output.png \
  --resolution 1K \
  --api-key {api_key}

​
Configuration Requirements

API Key:
  • Tool requires valid API key for Gemini 3 Pro Image
  • Configured during tool initialization
Workspace:
  • Must have write permissions to workspace directory
  • Requires generated/ subdirectory (created automatically)
Dependencies:
  • Python UV runtime
  • Image generation script installation
  • Network access to model API

​
Performance Considerations

Generation time by resolution:
  • 1K: ~5-10 seconds
  • 2K: ~10-20 seconds
  • 4K: ~20-40 seconds
File sizes:
  • 1K: ~500 KB - 2 MB
  • 2K: ~2 MB - 5 MB
  • 4K: ~5 MB - 15 MB
Generation times are approximate and depend on prompt complexity, API load, and network latency.

​
Best Practices

Prompt writing:
  • Be specific and descriptive
  • Mention style, mood, and setting
  • Enhancement handles technical details automatically
Good prompts:
  • “Mountain landscape at golden hour with dramatic clouds”
  • “Close-up portrait of elderly man with weathered face”
  • “Modern minimalist living room with natural light”
Avoid vague prompts:
  • “Nice picture” (too generic)
  • “Something cool” (no direction)
  • “AI art” (conflicts with enhancement style)
Resolution selection:
  • Use 1K for quick previews and web display
  • Use 2K for high-quality prints and presentations
  • Use 4K for maximum detail and professional use
Aspect ratio guide:
  • 1:1 - Social media posts, profile pictures
  • 3:4 - Portrait photography, print photos
  • 4:3 - Traditional displays, presentations
  • 9:16 - Mobile apps, Instagram stories
  • 16:9 - Desktop wallpapers, video thumbnails

Build docs developers (and LLMs) love

Get started for freeTalk to us