Analyze video file:
```bash
~/.claude/skills/gemini/webapi "Summarize this video" --file video.mp4
```

Analyze YouTube video:
```bash
~/.claude/skills/gemini/webapi "What are the key points?" --youtube "https://youtube.com/watch?v=..."
```

Analyze document:
```bash
~/.claude/skills/gemini/webapi "Summarize this report" --file report.pdf
```

Generate image:
```bash
~/.claude/skills/gemini/webapi "A sunset over mountains" --generate-image sunset.png
```

Edit image:
```bash
~/.claude/skills/gemini/webapi "Make the sky purple" --edit photo.jpg --output edited.png
```

Show thinking process:
```bash
~/.claude/skills/gemini/webapi "Solve step by step: What is 15% of 240?" --show-thoughts
```

<cli_options> | Option | Description | |--------|-------------| | --file, -f FILE | Input file (MP4, PDF, PNG, JPG, etc.) | | --youtube URL | YouTube video URL to analyze | | --generate-image FILE | Generate image and save to FILE | | --edit IMAGE | Edit existing image (use with --output) | | --output, -o FILE | Output file path for image operations | | --aspect RATIO | Aspect ratio for images (16:9, 1:1, 4:3, 3:4) | | --show-thoughts | Display model's thinking process | | --model MODEL | Model to use (default: gemini-3.0-pro) | | --json | Output response as JSON | | --retries N | Number of retries for transient failures (default: 3) | | --timeout SECS | Client initialization timeout in seconds (default: 120) | | --help, -h | Show help | </cli_options>

<example name="Video meeting summary">
  ```bash
  ~/.claude/skills/gemini/webapi "Create meeting notes: attendees, topics, decisions, action items" --file meeting.mp4
  ```
</example>

<example name="YouTube tutorial extraction">
  ```bash
  ~/.claude/skills/gemini/webapi "List all steps with timestamps" --youtube "https://youtube.com/watch?v=VIDEO_ID"
  ```
</example>

<example name="Current events (grounded)">
  ```bash
  ~/.claude/skills/gemini/webapi "What are the latest developments in AI this week? Search the web."
  ```
</example>

<example name="Image generation with aspect ratio">
  ```bash
  ~/.claude/skills/gemini/webapi "A cyberpunk cityscape at night" --generate-image city.png --aspect 16:9
  ```
</example>

<example name="Document analysis">
  ```bash
  ~/.claude/skills/gemini/webapi "Extract the key findings and recommendations" --file research_paper.pdf
  ```
</example>

<issue name="Transient/timeout errors">
  <symptom>Timeout, connection reset, or 5xx errors</symptom>
  <solution>The client automatically retries with backoff. If failures persist, try --retries 5 or --timeout 180. Check your network connection.</solution>
</issue>

<issue name="No images generated">
  <symptom>Script runs but no image output after retries</symptom>
  <solution>
    1. Rephrase prompt to be more explicit about wanting an image
    2. Some content may be filtered by safety systems
    3. Try simpler prompts first to verify the client works
    4. Check if the model returned text (may indicate why image failed)
  </solution>
</issue>

<issue name="Module not found">
  <symptom>ImportError for gemini_webapi</symptom>
  <solution>Run: cd ~/.claude/skills/gemini && source .venv/bin/activate && pip install -r requirements.txt</solution>
</issue>

<issue name="YouTube not working">
  <symptom>Cannot analyze YouTube video</symptom>
  <solution>Ensure YouTube extension is enabled in your Gemini web settings at gemini.google.com</solution>
</issue>

<issue name="No thinking shown">
  <symptom>--show-thoughts returns nothing</symptom>
  <solution>Thinking may be skipped for trivial queries. Try a more complex question, or the model may have used code execution instead.</solution>
</issue>

<issue name="Rate limiting">
  <symptom>Repeated failures with rate limit errors</symptom>
  <solution>Wait a few minutes before retrying. Consider spacing out requests. The client uses exponential backoff automatically.</solution>
</issue>

<important_notes> Always uses "Thinking with 3 Pro" mode for enhanced reasoning Google Search grounding is automatic - just ask about current events YouTube extension must be enabled in gemini.google.com settings Cookie-based auth - no API key needed, uses your Google account quota First run on macOS prompts for Keychain access </important_notes> </gemini_skill>