Wan Audio Direction

Wan 2.2 local T2V/I2V is usually silent picture (S2V is the speech graph). Wan 3.0 / 3.0 Prime generates voice + SFX + music in the same pass, on by default. This skill is the 3.0 soundtrack contract. Do not use MiniMax <d> tags or overall_soundscape: fields here.

When to Use

• Any Wan 3.0 / Prime generate
• Clip invented VO or BGM you never asked for
• Dialogue timing across a 10–30 s shot list

Don't use for: H3 (minimax_h3_audio_direction); LTX (ltx23_audio_visual_sync); 2.2 Lightning graphs unless you are on S2V.

Alibaba's three separate things

Write each as its own sentence. Mixing them into one "cinematic audio" clause is how 3.0 picks a random ballad and a random line.

1. Voice

Exact line + delivery + language/accent.