A cinematic orchestral instrumental piece — solo piano opening, full orchestra at the lift — paired with 6 painterly-cinematic dawn scenes from intimate empty room to mountain vista to peaceful resolution. No vocals, no lyrics, no text. The visuals carry the entire arc. Same playbook as City Light, with mode: "instrumental" flipped. 10 minutes wall, ~$4 cost.






create_media action=music, model=minimax-music with TWO inputs:
prompt = "cinematic orchestral, 90 BPM, swelling strings + french horn, no vocals" ← STYLE
lyrics_prompt = "[Instrumental]" ← tells the model NO vocals at all
Output: 60s purely instrumental mp3 → done in ~46s
⚠ BUG class fixed 2026-06-03: without explicit [Instrumental] the singer
literally sang the STYLE prompt as the lyrics. Now required to pass either
[Instrumental] or real lyrics in lyrics_prompt.
mode: "instrumental"
dawn-arriving-reel.mp4 · 1920×1080 · 57.7s · 20MB
Total: ~$3.83. Same per-second economics as the lyric version — the only delta is the Stage-5 overlay step, which is free (local Pillow + ffmpeg).
City Light and Dawn Arriving are the same playbook with mode flipped. Same pipeline, same caps, same cost. The mode toggle decides whether lyric typewriter overlays fire in Stage 5.