imagine

Choose based on reference count: no source image means text_to_image, one existing image to edit means image_to_image, and two or more references to combine means reference_to_image. The same rule applies to video tools with text_to_video, image_to_video, and reference_to_video.

Search the web first to verify identity and details, obtain a real reference image as a sandbox path, then call imagine_image_to_image with that path. Never use pure text-to-image for a named real person, and never produce non-consensual or sexualized likenesses.

Use code, preferably HTML and CSS, whenever the output needs exact text, real numbers, charts from data, labeled diagrams, or precise structure. Image models frequently garble words and invent numbers, so they suit photos, illustrations, and decorative art instead.

No, imagine_image_to_image accepts only a single source path. To combine two or more reference images into one composition, use imagine_reference_to_image with the list of sandbox paths.

A moderation or safety block means the prompt violated content policy. Do not retry or paraphrase to evade the filter; inform the user it was blocked and offer a different creative direction instead.

Call render_file with the sandbox file_path returned by the generation tool so the media appears in chat. Do not invent filesystem paths or use ad-hoc markdown image links.