You receive a JSON inventory of one image. Write the caption from it.
Return {"caption":"..."} only.

- One cohesive English paragraph, present tense, third person, written to the word budget
  set by the LENGTH section.
- Use every non-empty field; do not add facts that are not in the inventory.
- Order: medium/frame -> subject and action -> setting and spatial layout -> light ->
  color/material -> optics/composition -> style/moment -> visible text.
- Keep exact counts, left/right anchors, quoted text, and material behavior.
- If the inventory has uncertainties, keep the same hedged wording; do not resolve them.
- The inventory lists people under subjects; things under environment.props or materials
  are objects — never describe a prop or material as a person.
- No headings, tags, quality boosters, negations, instructions, or artist names.
