| name | create-image |
| description | Generate images from a text prompt and display them with prompt, status, and progress. Use for "make an image of…", "generate a picture of…", "imagine…", or visualizing a concept/mood. Triggers: generate image, make a picture, draw, imagine, create art, render an image, AI image, visualize this. |
| license | MIT |
| compatibility | Requires JarvisVR agent-backend with an image-generation provider configured; offline shows queued/placeholder state. |
| metadata | {"agent":"media-agent","category":"media","version":"1.0","author":"jarvisvr"} |
| allowed-tools | show_generated_image |
Create Image
Turn a prompt into one or more images shown in an image_gen_viewer, reflecting
the live generation lifecycle (queued → generating → done).
Steps
- Refine the prompt. Keep the user's intent; you may add concise style hints
("synthwave, high detail") but don't override their request.
- Spawn the viewer with
show_generated_image in queued/generating
state and a progress value.
- Stream progress with
holo.update as generation advances, then set
status:"done" with the resulting images.
- Speak a short confirmation; offer variations or a save.
Output
image_gen_viewer (show_generated_image, props per registry.json):
While generating:
{ "widget_type": "image_gen_viewer",
"transform": { "anchor": "world", "billboard": true },
"props": { "prompt": "a neon city skyline at night, synthwave",
"status": "generating", "progress": 0.4, "model": "diffusion-xl" },
"interactions": ["tap","grab","resize","drag","dwell"] }
When done (holo.update):
{ "object_id": "O_imggen",
"props": { "status": "done", "progress": 1.0,
"images": [ { "url": "https://cdn.jarvisvr.app/gen/neon1.png", "seed": 42 } ] } }
agent.speech: { "text": "Here's your synthwave skyline.", "final": true }
Edge cases
- No provider / generation fails → set
status:"error" with an error
message; don't spawn fake image URLs.
- Unsafe / disallowed prompt → decline briefly and offer a safe reframe.
- Multiple variations → return several
images; user taps select_image.
- Use as cover art / texture → hand the URL to
control-media-playback
(poster) or a model_viewer/image_board via the stage.
- Very slow gen → keep streaming
progress; let the user keep talking.
