Google Cloud Vertex AI Image Generator (ADC)#

This skill utilizes the Google Cloud SDK ( gcloud) and application credentials to call the Google Cloud Vertex AI generateContent API, generating images directly using model predictions.

Security & Privacy Notice#

• Process Execution: This skill invokes subprocess to orchestrate gcloud and curl for secure API interaction.
• Privilege & Access: It requires access to your environment variables (for authentication context) and local file read/write permissions to handle image prompt files and generate the output assets.
• Debug Artifacts: For error tracking and development transparency, full API responses may be written to the system-defined temporary directory. These artifacts are local to the machine running the skill.

Workflow#

1. Configure required environment variables or parameters:
   • GOOGLE_CLOUD_PROJECT: Your Google Cloud Project ID.
   • GOOGLE_CLOUD_LOCATION (Optional): Location/region (defaults to global).
   • GOOGLE_CLOUD_MODEL (Optional): Model name (defaults to gemini-3.1-flash-image-preview).
2. Construct/save prompt to a local text file.
3. Call the generation script.

Commands#

Generate Image (Default ADC)#

bash

python3 {baseDir}/scripts/image_harness.py --prompt-file <prompt_path> --out <output_path>

Advanced Usage#

• Specify custom project, location, and model:

bash

python3 {baseDir}/scripts/image_harness.py --prompt-file <prompt_path> --out <output_path> --project <project_id> --location <location> --model <model>

• Use user-auth token instead of ADC:

bash

python3 {baseDir}/scripts/image_harness.py --prompt-file <prompt_path> --out <output_path> --auth-mode user-token

Setup & Requirements#

• Google Cloud SDK ( gcloud CLI) installed and available in $PATH.
• Active authentication token:
  • Default ADC mode: gcloud auth application-default login
  • User-token mode: gcloud auth login
• The system must have curl installed to issue the API request.