Generate images and keep refining

Describe the image you want to create or change to Agent. Discuss directions, organize steps or run a Skill in the same conversation.

Generate a first round

1. Request images; default models and parameters are strongly recommended. State aspect ratio, resolution or intended use when specific.
2. Attach sketches, existing designs or references and wait until ready, or start with a text description.
3. Define subject, viewpoint, setting, this round's focus and fixed conditions.
4. Check inputs and parameters, send and follow progress in the original conversation.

Distinguish the base design from references: the base might define massing and entrances while another image supplies materials. Avoid conflicting roles.

Inspect and choose

Check atmosphere and details such as floor count, openings, silhouette, people and text. Compare multiple outputs in a frame. Use Show on canvas to locate a result; follow the prompt if it is on another Page.

Refine a specific image

Reference the chosen image and state what changes and what stays: “Keep the massing and viewpoint; change only the materials beside the entrance.” Remove unused attachments.

For cropping, extending or local edits, select the image and use the appropriate tool. Distinguish cropping existing pixels from generating new content, which may change details.

Regeneration creates a new result and does not guarantee unchanged details. Retain versions you need and identify the one to use next.

Image tools · References · Parameters · Documentation