Use the shared method. Output contract for MOSS-Image single-image generation:

- rewritten_prompt: one cohesive English paragraph, within the word budget set by the
  LENGTH section, present tense; medium
  first, then subject, setting, spatial layout, light, color/material, lens/composition.
  Action belongs to the body of the paragraph, never to sentence one. Do not name a mood:
  no sentence may end on an emotion word. Cohesive prose only: no headings, bullets, tags,
  or keyword piles.
- Do not add bounding boxes and do not force a layout. Layout is only for explicit text
  placement, posters, panels, or controlled object relations.
- Visible text: quote exactly in its original script, give placement and typography. If the
  user did not ask for text, do not invent signage.
- No quality boosters, artist names, negations, or process notes.
- wh_ratio: the user's ratio if given; otherwise choose from 3:2, 2:3, 1:1, 16:9, 9:16, 4:3,
  3:4. Do not write the ratio into the prose.
- If the user gives explicit pixel dimensions, keep them; if they are not multiples of 16 or
  conflict with the ratio, report the conflict in notes instead of silently changing them.
- Binding over length: attach counts, colors, materials, spatial relations, and text to the
  correct object; do not pad or add unattached adjectives.
- must_keep: 3-8 invariants. open_choices: the main creative decisions you made.
