duty-image

Image duty

Fulfil the image duty: turn a text prompt into an image file and hand back its path. Keep this simple and honest — it is a documented working path, not an automated generation service.

The path

1. Pick an image-capable model the session can reach.

   • Default: the gemini runtime. The gemini CLI ships with the gemini-runtime fitting. Use an image-generation model (e.g. an Imagen model) — the general text model (gemini-2.5-flash) does not generate images, so name an image model explicitly.
   • Alternative: an image MCP tool. If the running composition exposes an image-generation MCP tool, call that instead.

2. Send the prompt. Pass the user's description through to the chosen model. Example shape (adjust the model/flags to the installed CLI version):

   gemini generate-image --model <image-model> \
     --prompt "<the image description>" \
     --out ~/.garrison/files/<slug>.png
   

3. Save to the artifact store. Write the image under ~/.garrison/files with a descriptive name. The user opens it in Projects, under Garrison files. Return the saved path as the duty's output.

Honesty

• This duty depends on an image-capable model being configured and reachable. If none is (only text models are available), say so plainly and stop — do not fabricate a file or claim an image was produced.
• Do not build a heavy generation pipeline here. One prompt in, one saved image out, through whatever image-capable model the composition already provides.