You inventory one image for a reconstruction team. Return JSON only, no prose outside it.

{
 "format": {"medium":"","genre":"","orientation":"","aspect_ratio":""},
 "subjects": [{"kind":"person","role":"","count":0,"appearance":"","wardrobe":"","pose":"","gaze":"","expression":"","hands":"","feet":"","position":""}],
 "environment": {"place":"","time":"","weather":"","foreground":"","midground":"","background":"","props":[]},
 "camera": {"shot":"","height":"","angle":"","lens":"","dof":"","composition":"","cropping":""},
 "light": {"sources":[],"direction":"","quality":"","color_temperature":"","contrast":"","shadows":"","highlights":""},
 "palette": {"dominant":[],"accent":[],"skin_anchor":""},
 "materials": [{"surface":"","behavior":""}],
 "text": [{"text":"","where":"","typography":""}],
 "style": {"medium":"","era":"","treatment":""},
 "moment": {"action_phase":"","atmosphere":""},
 "confidence": {"subjects":0.0,"count":0.0,"color":0.0,"material":0.0,"space":0.0,"text":0.0},
 "uncertainties": []
}

Rules: visible only; exact counts; no brands unless legible; no quality adjectives; unknown
fields stay empty strings or empty arrays; do not guess identities or backstory.

kind: "person" only for a human being (or a character the work explicitly depicts as a
person). A cup, window, lamp, vehicle, food, tool, wall, fabric or animal statue is
"object" and belongs in environment.props or materials, never in subjects. Never give a
non-person an age, gender, hair, skin or expression.
At most 12 subjects, 32 props, 16 materials and 24 text items. List each distinct subject
once; never repeat the same label with a new position.
