Nano Banana

⚠️ Always run source /etc/profile first — environment variables are stored there and won't be available otherwise.

If not set, direct the user to Google AI Studio to obtain a key, then:

echo 'export GEMINI_API_KEY="your_key_here"' >> /etc/profile

Check Dependencies

pip show google-genai pillow 2>&1 | grep -E "^Name|not found"

If missing:

pip install google-genai pillow

Model Selection

Default to gemini-3.1-flash-image-preview unless the user explicitly requests another version.

⚠️ Known API Pitfalls

1. Aspect ratio / resolution config

❌ Wrong (old API, throws AttributeError):

# types.ImageGenerationConfig does not exist!
image_generation_config=types.ImageGenerationConfig(aspect_ratio="16:9")

✅ Correct:

config=types.GenerateContentConfig(
    response_modalities=["IMAGE"],
    image_config=types.ImageConfig(
        aspect_ratio="16:9",  # options: 1:1, 4:3, 3:4, 16:9, 9:16
        image_size="2K",      # options: 1K, 2K (default: 1K)
    ),
)

2. Environment variable loading

❌ Running python3 script.py directly may not have the API key
✅ Always prefix with source /etc/profile && python3 script.py

3. Saving images

part.as_image() returns a PIL Image object — call .save(path) directly. No need to handle base64 manually.

Bundled Scripts

Run any script with:

source /etc/profile && python3 <skill-path>/nano-banana/scripts/<script>.py [args]

📄 gen.py — Text to Image

Usage:  gen.py "prompt" [output_path] [aspect_ratio] [resolution]
Example: gen.py "a panda drinking tea in a bamboo forest" /var/minis/attachments/out.png 1:1 2K

📄 edit.py — Image Editing (image + prompt → image)

Usage:   edit.py <input_image> "edit instruction" [output_path]
Example: edit.py /var/minis/attachments/photo.jpg "add a wizard hat to the cat" /var/minis/attachments/edited.png

📄 batch.py — Batch Generation (multiple prompts → multiple images)

Edit the TASKS list at the top of the script, then run it.

Step-by-Step

When the user makes an image generation request:

1. Understand the request — text-to-image, image editing, or batch?
2. Check API key — source /etc/profile && echo $GEMINI_API_KEY | head -c 10
3. Verify scripts exist — ls <skill-path>/nano-banana/scripts/
4. If scripts are missing, use file_write to recreate them from the source in this skill
5. Run the appropriate script:
   • Text-to-image: source /etc/profile && python3 <skill-path>/nano-banana/scripts/gen.py "prompt" /var/minis/attachments/out.png 16:9 2K
   • Image editing: source /etc/profile && python3 <skill-path>/nano-banana/scripts/edit.py <image_path> "instruction" /var/minis/attachments/out.png
   • Batch: edit TASKS in scripts/batch.py, then run it
6. Display the result — ![description](minis://attachments/out.png)

Prompt Tips

• Describe a scene, don't just stack keywords — "An elderly Japanese potter focused at the wheel, soft afternoon light through a studio window, photorealistic" beats "Japan pottery old man realistic"
• Specify style explicitly — for photography add camera details ("50mm lens, shallow depth of field"); for illustration name the style ("Ghibli-style watercolor")
• Text in images — spell out exactly what text should appear; the model renders it more accurately when it's explicit
• Iterate with edit.py — make small adjustments on an existing image rather than regenerating from scratch
• Aspect ratio guide — 16:9 for banners/social media, 1:1 for avatars/covers, 9:16 for phone wallpapers, 3:4 for portrait cards

Troubleshooting

References

• Official docs: https://ai.google.dev/gemini-api/docs/image-generation
• Get an API key: https://aistudio.google.com/apikey