Convert the following T2I image generation prompt into a final scene description for an image generation model.

CRITICAL RULES:
1. The image generation model does NOT know any character names, location names, or proper nouns
2. Replace character/object IDs (C##O##, P##, etc.) with "the character/object shown in image N" format — ONLY when the ID has a matching Reference image
3. Animals, unnamed people (extras, bystanders), and any subject WITHOUT a Reference image ID: keep the original text description as-is. Do NOT map them to any Reference image
4. Do NOT describe character appearance in the scene body for characters that HAVE reference images — their appearance is already defined there
5. In the scene body, describe ONLY actions, positions, poses, and spatial relationships for referenced characters. For non-referenced subjects (animals, extras), keep full visual descriptions
6. Translate to English — EXCEPT: proper nouns (place names, business names, etc.) should preferably stay in original language. Also keep any expression in original language where visually natural or where English translation would be awkward
7. Remove ALL marker brackets: [[]], [:]
8. Do NOT include camera angles or shot types (no "close-up", "wide shot", "medium shot")
9. Do NOT include "Photorealistic cinematic still." — it will be added separately
10. Do NOT include reference image descriptions or generation instructions — output ONLY the scene body text

ID → REFERENCE IMAGE MAPPING:
{ref_roles_text}

REFERENCING RULES:
- Characters: use "the character shown in image N" (never describe their face, hair, outfit, or body)
- Objects: use "the object shown in image N"
- You may add a brief action-only parenthetical: "the character shown in image 1 (lunging forward)" — but NO appearance words

PHANTOM PHRASE PROHIBITION (strict):
- NEVER produce the phrases "from the reference", "from Reference image N", "from a reference image", or any "from … reference" wording in the output
- These phrases trigger a downstream phantom-reference validator and will fail the render
- Use ONLY "shown in image N" / "in image N" / "depicted in image N" when referring to a numbered reference
- For generic instructions (do not copy poses, do not duplicate compositions, etc.), do NOT include them in the scene body — they belong to a separate instruction block

World/style context to incorporate as visual atmosphere (NOT as named locations):
{style_context}

Original T2I prompt to convert:
{t2i_prompt}

Output ONLY the scene body text. No reference image descriptions, no generation instructions, no explanations.