| name | nano-image-generate |
| description | Generate images using Nano Banana (Flash) or Nano Banana Pro. Use 'flash' for speed/efficiency and 'pro' for high quality, text rendering, and complex prompt adherence. Triggers include 'generate image', 'create logo', 'fast image', 'high quality image'. |
| metadata | {"version":"1.1"} |
Nano Image Generator
Generate images using Nano Banana Pro (Gemini 3 Pro Preview) for any visual asset needs. Supports reference images for style transfer and character consistency.
Quick Start
[!IMPORTANT]
Path Check: Before running, localize the scripts/generate_image.py file. If running this skill from another project, use the absolute path to the script or copy it to your project.
python scripts/generate_image.py "A friendly robot mascot waving"
python scripts/generate_image.py "Same style, new content" --ref ./reference.jpg --output ./new.png
Model Selection
Choose the model based on user intent:
| Model | ID | Use Case | Keywords |
|---|
| Pro | gemini-3-pro-image-preview | Quality & Detail. Default choice. Best for text, complex prompts, logos, and high-fidelity assets. | "high quality", "logo", "text", "detail", "complex" |
| Flash | gemini-2.5-flash-image | Speed & Efficiency. Best for rapid prototyping, large batches, or when low latency is critical. | "fast", "quick", "draft", "rapid" |
Script Usage
python scripts/generate_image.py <prompt> --output <path> [options]
Required:
prompt - Image description
Options:
--output, -o - Output file path. Optional. Defaults to ./generated/<slug>.png.--model, -m - Model selection: pro (default) or flash--key, -k - Gemini API Key (optional)--aspect, -a - Aspect ratio (default: 1:1)--size, -s - Resolution: 1K, 2K (default), 4K--ref, -r - Reference image (max 14)
Examples
High Quality Logo (Pro - Default):
python scripts/generate_image.py "Futuristic company logo with text 'NANO'" --output ./logo.png
Quick Prototype (Flash):
python scripts/generate_image.py "Draft sketch of a cat" --model flash --output ./cat_draft.png
Reference Images
Gemini supports up to 14 reference images for:
Style Transfer
Keep the visual style (colors, textures, mood) from a reference:
python scripts/generate_image.py "New scene with mountains, same visual style as reference" \
--ref ./style-reference.jpg --output ./styled-mountains.png
Character Consistency
Maintain character appearance across multiple images:
python scripts/generate_image.py "Same character now in a forest setting" \
--ref ./character.png --output ./character-forest.png
Multi-Image Fusion
Combine elements from multiple references:
python scripts/generate_image.py "Combine the style of first image with subject of second" \
--ref ./style.png --ref ./subject.png --output ./combined.png
Serial Image Generation (Batch Workflow)
For generating a series with consistent style:
- Generate first image
- Use first image as
--ref for subsequent images
- Each new image inherits the established style
python scripts/generate_image.py "Tech knowledge card cover" -o ./01-cover.png
python scripts/generate_image.py "Card 2 content, same style" --ref ./01-cover.png -o ./02-card.png
python scripts/generate_image.py "Card 3 content, same style" --ref ./01-cover.png -o ./03-card.png
Directory Structure
To keep things organized, this skill uses a standard directory structure:
./generated/ - Default output location. Images are automatically saved here if no path is provided../assets/ - Recommended for storing reference images.
Workflow
-
Verify Output Directory & Script Path:
- Ensure the output directory exists (e.g.,
mkdir -p generated) or is writable.
- Confirm the location of
generate_image.py.
-
Craft effective prompts - Be specific and descriptive:
- Include style: "flat design", "3D rendered", "watercolor", "minimalist"
- Include context: "for a mobile app", "website hero image"
- Include details: colors, mood, composition
- For references: mention "same style as reference" or "keep the visual style"
-
Choose appropriate settings:
- Icons/logos →
--aspect 1:1
- Banners/headers →
--aspect 16:9 or 21:9
- Mobile screens →
--aspect 9:16
- Xiaohongshu cards →
--aspect 3:4
- Photos →
--aspect 3:2 or 4:3
Examples
App icon:
python scripts/generate_image.py "Minimalist flat design app icon of a lightning bolt, purple gradient background, modern iOS style" \
--output ./assets/app-icon.png --aspect 1:1
Marketing banner:
python scripts/generate_image.py "Professional website hero banner for a productivity app, abstract geometric shapes, blue and white color scheme" \
--output ./public/images/hero-banner.png --aspect 16:9
Xiaohongshu knowledge card:
python scripts/generate_image.py "Tech knowledge card, dark blue purple gradient, neon cyan accents, code block style, Chinese text '标题'" \
--output ./xiaohongshu/card.png --aspect 3:4
Style transfer:
python scripts/generate_image.py "Transform this photo into watercolor painting style" \
--ref ./photo.jpg --output ./watercolor.png
Character in new scene:
python scripts/generate_image.py "Same character from reference, now sitting in a cafe, warm lighting" \
--ref ./character.png --output ./character-cafe.png --aspect 3:2
Prompt Tips
- Be specific - "A red apple on a wooden table" vs "an apple"
- Include style - "in the style of pixel art" or "photorealistic"
- Mention purpose - "for a children's book" affects the output style
- Describe composition - "centered", "rule of thirds", "close-up"
- Specify colors - Explicit color palettes yield better results
- Reference prompts - Use "same style as reference", "keep the visual aesthetic", "match the color palette"
- Avoid - Don't ask for complex text in images (use overlays instead)
Limitations
- Maximum 14 reference images per request
- Text rendering may be imperfect (better to overlay text separately)
- Very specific brand logos may not reproduce exactly
