Image Generation

Generate images from text descriptions using Google's Gemini model via the generate_image tool.

Setup

1. Enable image generation in Settings UI or set NOMOS_IMAGE_GENERATION=true
2. Get a Gemini API key from https://aistudio.google.com/apikey
3. Set GEMINI_API_KEY in Settings or environment

Tool: generate_image

Parameters:

• prompt (required) — Detailed description of the image to generate
• output_path (optional) — File path to save the image (defaults to temp directory)

Writing Effective Prompts

Good image prompts are specific about:

• Subject: What is in the image (person, object, scene)
• Style: Photorealistic, watercolor, oil painting, digital art, pencil sketch, etc.
• Composition: Close-up, wide shot, aerial view, symmetrical, rule of thirds
• Lighting: Golden hour, studio lighting, dramatic shadows, soft diffused light
• Colors: Warm palette, monochrome, vibrant, muted pastels
• Mood: Serene, dramatic, playful, mysterious

Prompt Examples

Simple:

A golden retriever sitting in a field of sunflowers at sunset

Detailed:

A photorealistic close-up of a steaming cup of coffee on a weathered wooden table, morning light streaming through a window, shallow depth of field, warm color palette

Artistic:

An Art Nouveau illustration of a woman surrounded by flowing botanical patterns, muted earth tones with gold accents, decorative border, inspired by Alphonse Mucha's style

Technical:

An isometric 3D rendering of a modern smart home cutaway showing interior rooms, clean minimal style, soft shadows, pastel color scheme on white background

Capabilities

• Photorealistic images
• Illustrations and concept art
• Logos and icons
• Diagrams and infographics (with text)
• Image editing via text description (describe modifications)
• Multiple art styles (watercolor, oil, digital, pixel art, etc.)

Output

• Images are saved as PNG (default), JPEG, or WebP
• Default save location is the system temp directory
• Specify output_path to save to a specific location (e.g., current working directory)

Tips

• More detail in prompts produces better results
• Specify image dimensions or aspect ratio in the prompt if needed
• For consistent style across multiple images, include the same style descriptors
• The model may also return text alongside the image with notes about what it generated