gemini-multimodal

gemini-multimodal

Gemini multimodal skill for Claude Code. Video, PDF, image analysis & generation via browser cookies - no API key required.

Installation

npx gemini-multimodal

This installs the skill to ~/.claude/skills/gemini/ and sets up the Python environment automatically.

Prerequisites:

• Python 3.8+
• Chrome logged into gemini.google.com
• On macOS, allow Keychain access when prompted (first run)

Features

Capability	Description
Text Queries	Complex reasoning with "Thinking with 3 Pro" mode
Video Analysis	Upload MP4 files for summarization, timestamps, insights
YouTube Analysis	Analyze videos via URL (uses YouTube extension)
Document Analysis	PDF and document Q&A
Image Analysis	Describe, OCR, analyze uploaded images
Image Generation	Create images from text prompts
Image Editing	Modify images with natural language
Google Search	Automatic grounding for current information

How It Works

User Request → webapi CLI → gemini-webapi → Gemini Web (cookies) → Response

Authentication uses Chrome browser cookies - no API key needed. Just be logged into gemini.google.com.

Usage

Text Queries

# Complex reasoning (Thinking with 3 Pro)
webapi "Explain the implications of quantum computing for cryptography"

# Show thinking process
webapi "Solve step by step: What is 15% of 240?" --show-thoughts

File Analysis

# Video analysis
webapi "Summarize this video with timestamps" --file meeting.mp4

# Document analysis
webapi "Extract key findings" --file research.pdf

# Image analysis
webapi "What's in this image?" --file photo.png

YouTube Analysis

webapi "What are the main points discussed?" --youtube "https://youtube.com/watch?v=VIDEO_ID"

Requires YouTube extension enabled in gemini.google.com settings.

Image Generation

# Generate image
webapi "A cyberpunk cityscape at night" --generate-image city.png

# With aspect ratio
webapi "Mountain landscape" --generate-image landscape.png --aspect 16:9

# Edit existing image
webapi "Make the sky purple" --edit photo.jpg --output edited.png

Current Information (Grounded)

webapi "What are the latest AI news this week? Search the web."

Google Search grounding is automatic when queries need current information.

CLI Options

Option	Description
--file, -f FILE	Input file (MP4, PDF, PNG, JPG, etc.)
--youtube URL	YouTube video URL
--generate-image FILE	Generate and save image
--edit IMAGE	Edit image (with --output)
--output, -o FILE	Output path for images
--aspect RATIO	Aspect ratio (16:9, 1:1, 4:3, 3:4)
--show-thoughts	Display thinking process
--model MODEL	Model to use (default: gemini-3.0-pro)
--json	JSON output
--help, -h	Show help

File Structure

gemini/
├── SKILL.md           # Claude Code skill definition
├── README.md          # This file
├── requirements.txt   # Python dependencies
├── .venv/             # Virtual environment
├── webapi             # Bash wrapper
└── webapi.py          # Python implementation

Troubleshooting

"Error initializing client"

• Log into gemini.google.com in Chrome
• On macOS, allow Keychain access when prompted

"No images generated"

• Rephrase prompt, some content is filtered
• Be more explicit about what you want

"Module not found"

• Activate venv: source .venv/bin/activate
• Install deps: pip install -r requirements.txt

"YouTube not working"

• Enable YouTube extension in gemini.google.com settings

License

MIT