Antigravity Image Generation

Use the internal Google Antigravity sandbox to generate images with the gemini-3-pro-image model. This skill bypasses standard API restrictions by impersonating the VS Code plugin environment.

Usage

Generate an image from a prompt:

python3 {baseDir}/scripts/generate_image_antigravity.py --prompt "A futuristic cityscape with flying cars" --filename "city.png"

How it works

1. Authentication: Automatically retrieves OAuth tokens and Project ID from the local auth-profiles.json.
2. Impersonation: Uses the specific antigravity User-Agent and internal sandbox headers.
3. Reliability: Includes automatic retry logic for 503 Service Unavailable errors.

Parameters

• --prompt / -p: Text description of the image to generate.
• --filename / -f: Path where the resulting PNG will be saved.

Notes

• The script prints a MEDIA: path which OpenClaw uses to automatically upload the image to the chat channel.
• Requires an active Antigravity session/login in the environment.
• See internal-api.md for technical details on the impersonated endpoint.