Get detected items + scene narrative for an image

Returns the latest completed detection scan for an image you own: every detected item with its name, category, texture, hex colors, and bounding box, plus a 150-word scene narrative and overall style.

Bounding boxes are normalized to a 0-1000 square regardless of pixel size — multiply by width/1000 and height/1000 to get pixels. Items form a hierarchy via parent_id (a sleeve's parent is the jacket).

Every item has an id, unique within the scan and the same on every read of it (scan_id); a rescan numbers its items afresh. label is the name, or, when several items share a name, the name plus where this one sits in the picture: "person (left)", "cup (2nd from left)" (the picture's left, not the subject's). To remove one of two people, pass its id to POST /v1/restyle (action: "remove", items: [{ "id": 2 }], optionally scan_id).

Free. If the image has never been scanned (404), trigger one with POST /v1/scans — though every generation and import is scanned automatically, so this mostly matters for very old images.