MiniMax H3 Audio Direction

H3 always generates 32 kHz stereo in the same pass as picture. There is no silent default. Unspecified audio is the #1 first-week failure (speech-like garbage, unrequested room tone, invented VO).

Picture dialect lives in minimax_h3_prompt_engineering_master. This file is the soundtrack contract from the official Base + Full-Reference guides.

When to Use

• Any H3 T2VA / I2VA / FL2VA / L2VA / Ref2VA
• Lip-sync, VO, music-video, product foley
• User says the clip "sounds weird" or "talking over nothing"

Don't use for: Flux stills; LTX audio (ltx23_audio_visual_sync); Wan 3.0 three-layer prose (wan_audio_direction).

Three disjoint layers

Keep them separate. Do not copy a spoken line into the soundscape.