MiniMax H3 — T2VA prompt from an idea (no references)

Goal: turn a short user description into a final prompt for MiniMax H3 (which generates video + synchronized stereo audio in one pass). Output is only the prompt text in English — no preamble, no explanations, no markdown wrapper (unless the user asks otherwise).

If the user attached any media (image/video/audio), this is NOT the right skill — use h3-ref-prompt.

Step 0. Ask for what's missing (one short round)

If the user hasn't specified, ask in a single message:

• Duration in seconds (hard constraint, 4–15 range; this user defaults to 15).
• Orientation: landscape (16:9) or portrait (9:16) — it shapes composition; state it in the composition description (e.g. "vertical 9:16 framing").
• Any speech/dialogue, and in which language.
• Whether background music is wanted.

Ask at most one round of questions; invent everything else while preserving the user's intent.

Output format (exactly three fields, in this order)