Gemini Text Generation

When to Use This Skill

• Generate any type of text content (blogs, emails, code, stories)
• Process images with text descriptions or analysis
• Perform complex reasoning requiring step-by-step thinking
• Get structured JSON outputs for data processing
• Access real-time information via Google Search
• Apply specific personas or behavior patterns
• Combine text generation with other Gemini skills (images, TTS, embeddings)

Available Scripts

scripts/generate.py

• Any text generation task
• Multimodal prompts (text + image)
• Complex reasoning requiring thinking mode
• Structured JSON output requirements
• Real-time information needs (grounding)
• Custom system instructions/personas

prompt

"Explain quantum computing"

--model

-m

gemini-3-flash-preview

--system

-s

"You are a helpful assistant"

--thinking

-t

--json

-j

--grounding

-g

--image

-i

photo.png

--temperature

0.7

--max-tokens

1000

Workflows

Workflow 1: Basic Text Generation

python scripts/generate.py "Explain quantum computing in simple terms"

• Best for: Simple content creation, explanations, summaries
• Model:

  gemini-3-flash-preview

  (default, fast)

Workflow 2: With System Instruction (Persona)

python scripts/generate.py "How do I read a file in Python?" --system "You are a helpful coding assistant"

• Best for: Domain-specific tasks, expert personas, consistent tone
• Use when: You need specific behavioral constraints

Workflow 3: Complex Reasoning (Thinking Mode)

python scripts/generate.py "Analyze the ethical implications of AI in healthcare" --thinking

• Best for: Complex analysis, step-by-step reasoning, multi-step problems
• Use when: Task requires careful consideration and logical progression

Workflow 4: Structured JSON Output

python scripts/generate.py "Generate a user profile object with name, email, and preferences" --json

• Best for: Data extraction, structured data generation, API responses
• Output: Valid JSON ready for parsing
• Note: Prompt must clearly request JSON structure

Workflow 5: Real-Time Information (Grounding)

python scripts/generate.py "Who won the latest Super Bowl?" --grounding

• Best for: Current events, news, factual information after training cutoff
• Output: Response + grounding sources with citations
• Use when: Accuracy of current information is critical

Workflow 6: Multimodal (Image Analysis)

python scripts/generate.py "Describe what's in this image in detail" --image photo.png

• Best for: Image captioning, visual analysis, image-based Q&A
• Requires: Image file in PNG or JPEG format
• Combines well with: gemini-files for file upload

Workflow 7: Content Creation Pipeline (Batch + Text + TTS)

# 1. Create batch requests (gemini-batch skill)
# 2. Generate content
python scripts/generate.py "Create a 500-word blog post about sustainable energy"
# 3. Convert to audio (gemini-tts skill)

• Best for: High-volume content production, podcasts, audiobooks

Parameters Reference

Model Selection

gemini-3-flash-preview

gemini-3-pro-preview

gemini-2.5-flash

gemini-2.5-pro

Temperature Settings

Thinking Budget

Output Interpretation

Standard Text Output

• Plain text response ready for use
• Check for truncation if max-tokens was set
• May include markdown formatting

JSON Output

• Valid JSON object (use

  --json

  flag)
• Parse with:

  import json; data = json.loads(output)

• Verify structure matches your requirements
• Handle potential parsing errors

Grounded Response

--grounding

1. Main response text
2. "--- Grounding Sources ---" section
3. List of sources with titles and URLs

Thinking Mode Output

• May include reasoning steps before final answer
• Longer response times due to thinking process
• Better for tasks requiring careful analysis

Common Issues

"google-genai not installed"

pip install google-genai

"API key not set"

export GOOGLE_API_KEY="your-key-here"
# or
export GEMINI_API_KEY="your-key-here"

"Model not available"

• Check model name spelling
• Verify API access for selected model
• Try

  gemini-3-flash-preview

  (most available)

JSON parse errors

• Ensure prompt explicitly requests JSON structure
• Check output for JSON formatting
• Consider using system instruction: "You always respond with valid JSON"

Image file not found

• Verify image path is correct
• Use absolute paths if relative paths fail
• Supported formats: PNG, JPEG

Response truncated

• Increase

  --max-tokens

  value
• Break task into smaller requests
• Use pro models with higher token limits

Best Practices

Performance Optimization

• Use flash models for speed, pro for quality
• Lower temperature (0.0-0.3) for deterministic outputs
• Set appropriate max-tokens to control costs
• Use thinking mode only for complex tasks

Prompt Engineering

• Be specific and clear in your prompts
• Use system instructions for consistent behavior
• Include examples in prompts for better results
• For JSON: specify exact structure in prompt

Error Handling

• Wrap script calls in try-except blocks
• Validate JSON output before parsing
• Handle network timeouts with retries
• Check API quota limits for batch operations

Cost Management

• Use flash models when possible (lower cost)
• Limit max-tokens for simple queries
• Cache results for repeated queries
• Use batch API for high-volume tasks

Related Skills

• gemini-image: Generate images from text
• gemini-tts: Convert text to speech
• gemini-embeddings: Create vector embeddings for semantic search
• gemini-files: Upload files for multimodal processing
• gemini-batch: Process multiple requests efficiently

Quick Reference

# Basic
python scripts/generate.py "Your prompt"

# Persona
python scripts/generate.py "Prompt" --system "You are X"

# Thinking
python scripts/generate.py "Complex task" --thinking

# JSON
python scripts/generate.py "Generate JSON" --json

# Search
python scripts/generate.py "Current event" --grounding

# Multimodal
python scripts/generate.py "Describe this" --image photo.png

Reference

• See

  references/models.md

  for detailed model information
• Get API key: https://aistudio.google.com/apikey
• Documentation: https://ai.google.dev/gemini-api