MiniMax H3 — full-reference prompt (with reference media)

Goal: turn the user's idea and attached media into a six-section full-reference brief for MiniMax H3. Output is only the brief text in English (original language survives only inside <d> tags and in on-screen text). If there is NO media — use the h3-t2v-prompt skill instead.

Step 0. Inputs

• List the media and their roles (ask the user if unclear): first-frame anchor / last-frame anchor / keyframe / general reference (character, scene, style, voice timbre, soundtrack, source video).
• Media order is semantic: labels are numbered by input order, independently per category — <Picture 1..N>, <Video 1..N>, <Audio 1..N>. Never rename, skip, renumber, or reorder.
• Ask in one message what's missing: duration (4–15s, this user defaults to 15), orientation (16:9 / 9:16), speech and its language.
• Audio can never be the only media — only alongside an image or video.

Output format (exactly six sections, in this order)

subject_definitions:
...

summary:
...