Image Prompting Best Practices

When helping users generate or edit images with nanobanana, apply these guidelines.

Prompt Structure for Generation

A strong image prompt follows this pattern: [Subject] + [Style] + [Composition] + [Lighting/Mood] + [Details]

Example: "A cozy coffee shop interior, watercolor illustration style, wide angle view, warm golden lighting, with plants on shelves and a cat sleeping on a chair"

Model Selection Guide

nanobanana supports three Gemini models via the NANOBANANA_MODEL env var:

Model	Best For
gemini-2.5-flash-image (default)	Fast generation, prototyping, high-volume work
gemini-3-pro-image-preview	Complex prompts, text rendering in images, high quality
gemini-3.1-flash-image-preview	Latest features, advanced capabilities

Recommend model changes when appropriate:

• User needs text in the image -> suggest Pro
• User is iterating rapidly -> stick with Flash (default)
• User wants highest quality for final output -> suggest Pro

Editing Best Practices

When using edit_image or continue_editing:

• Be specific about what to change: "Make the sky more orange" not "improve the colors"
• Reference specific areas: "Add a tree in the bottom-left corner"
• For style transfer, use reference images via the referenceImages parameter
• Each edit creates a new file -- the original is always preserved

Constraints

• Prompts over 10,000 characters will be rejected
• Image files must be under 20MB
• Supported input formats: JPEG, PNG, WebP, GIF
• Output is always PNG
• File paths must resolve within $HOME or $TMPDIR (security constraint)
• Images are saved to ~/nanobanana-images/

Iterative Workflow

The most effective image workflow is:

1. Generate a base image with a detailed prompt
2. Use continue_editing for incremental refinements
3. Each iteration should address ONE specific change
4. If the result diverges too far, start fresh with generate_image