image-generator

image-generator

Generate images using Gemini image models (default: gemini-3-pro-image-preview, Nano Banana Pro class).

Auth (preferred): set GEMINI_API_KEY or GOOGLE_API_KEY in the agent .env (Google AI Studio).
Fallback: Vertex AI via GCP_PROJECT_ID / VERTEX_AI_PROJECT_ID / GOOGLE_CLOUD_PROJECT plus Application Default Credentials.

When to use

• The user asks for a new image, illustration, diagram, or visual asset
• You need a PNG saved under ./generated-media/images/ for follow-up or display

How to run

Shell commands start in workspace root. The trusted generator script is in the separate read-only skills root from list_skills; generated output remains workspace-relative.

1. Confirm list_skills includes image-generator and note skills_path.
2. Run the generator with run_command and direct python3 argv (do not wrap in bash -c — that can pick a system Python without google-genai):

{
  "argv": [
    "python3",
    "<skills_path>/image-generator/generate_image.py",
    "--prompt",
    "A minimalist logo of a monkey astronaut, flat vector style",
    "--aspect-ratio",
    "16:9"
  ],
  "timeout": 180
}

Replace <skills_path> with the skills_path returned by list_skills (skills).

Supported --aspect-ratio values: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9.

3. Parse JSON stdout from the command result:

   • Success: {"ok": true, "path": "./generated-media/images/<uuid>.png", "prompt": "...", ...}
   • Failure: {"ok": false, "error": "..."}

4. Call load_file with the returned path only. That emits an ImageBlock to the chat UI (and loads pixels into model context on vision-capable providers):

load_file(path="<path from script>")

Do not use read_file on the PNG — that is for text files only.

5. Caption the image using the --prompt you submitted (or the echoed prompt field). Do not invent a different subject from memory or earlier chats. On text-only providers you will not see pixels — still describe from this turn's prompt.

Notes

• Output files are written under ./generated-media/images/ (allowed by run_command policy).
• Do not paste raw base64 in chat; always use load_file after generation.
• On failure, report the error field and do not call load_file. Never glob for **/*.png and load an arbitrary older file as a substitute.
• Do not use the task tool for this skill — run the steps above yourself.
• python3 / python in run_command use the gateway interpreter (which includes google-genai via monkeybot). Do not run pip install / uv pip install via run_command — installs are blocked by policy.